NISTIR 8034 Fingerprint Vendor Technology Evaluation Craig Watson Gregory Fiumara Elham Tabassi Su Lan Cheng Patricia Flanagan Wayne Salamon http://dx.doi.org/10.6028/NIST.IR.8034 NISTIR 8034 Fingerprint Vendor Technology Evaluation Craig Watson Gregory Fiumara Elham Tabassi Su Lan Cheng Patricia Flanagan Wayne Salamon Information Access Division Information Technology Laboratory This publication is available free of charge from: http://dx.doi.org/10.6028/NIST.IR.8034 December 2014 U.S. Department of Commerce Penny Pritzker, Secretary National Institute of Standards and Technology Willie May, Acting Under Secretary of Commerce for Standards and Technology and Acting Director Fingerprint Vendor Technology Evaluation Evaluation of Fingerprint Matching Algorithms NIST Interagency Report 8034 Craig Watson • Gregory Fiumara • Elham Tabassi Su Lan Cheng • Patricia Flanagan • Wayne Salamon 18 December 2014 iv FPVTE – F INGERPRINT M ATCHING Acknowledgements . Sponsors: The authors would like to thank the sponsors of this project: the Department of Homeland Security (DHS) and the Federal Bureau of Investigation (FBI). . Data Providers: The authors would like to thank all the organizations that have shared biometric data with NIST for this and other evaluations. This data has proven invaluable for benchmarking and advancement of biometric matching technologies through these evaluations. Data used in FpVTE came from the FBI, DHS, Los Angeles County Sheriff’s Department (LACNTY), Arizona Department of Public Safety (AZDPS), and Texas Department of Public Safety (TXDPS). . Participants: The authors would also like to thank the participants for their time and contribution to this evaluation. We know that preparing the software for the evaluation is not a trivial task and the validation process can be stressful when problems occur. Thank you for your time and participation as this evaluation would not be possible without it. Disclaimer Certain commercial equipment, instruments, or materials are identified in this paper in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by the National Institute of Standards and Technology, nor is it intended to imply that the materials or equipment identified are necessarily the best available for the purpose. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING v Contents Acknowledgements iv Disclaimer iv Executive Summary Caveats xiii xv Release Notes xvi 1 Introduction 1 2 History and Motivation 3 2.1 NIST Biometric Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 4 5 6 Data 5 3.1 Classes of Participation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 3.2 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 3.3 Evaluation Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.4 “Size” of the Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Experiment and Test Protocol 10 4.1 API Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.1.1 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.1.2 Finalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.1.3 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.2 Test Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4.3 Biometric Evaluation Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4.4 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 4.5 Pre-Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 4.6 Receiving Submissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Two-Stage Matching 16 5.1 Enrollment Set Partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 5.2 Identification — Stage One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 5.3 Identification — Stage Two . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Metrics 18 6.1 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 6.2 DET Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 6.3 Failure to Extract or Match a Template . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 6.4 Computational Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 6.5 Ground Truth Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology vi 7 8 9 FPVTE – F INGERPRINT M ATCHING Accuracy Results 22 7.1 Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 7.1.1 Single-Index Finger Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 7.1.2 Two-Index Finger Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 7.2 Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 7.3 Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Accuracy/Search Time Tradeoff 41 8.1 Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 8.2 Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 8.3 Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Accuracy Computational Resources Tradeoff 61 9.1 Storage and Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 9.1.1 Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 9.1.2 Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 9.1.3 Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Processing Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 9.2 10 Ranked Results 78 11 How Many Fingers are Needed 82 12 FpVTE 2003 Comparison 83 13 Lessons Learned for Large-Scale Testing 86 14 Way Forward 88 14.1 Related Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 14.1.1 Forensic Palmprint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 14.1.2 Mobile Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 14.1.3 Cross-Comparison of Modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 References 89 A Individual Participant FNIR Plots 90 B Combined Class DETs and CMCs 109 C Accuracy Time Tradeoff Detailed Tables with Median Values 114 D Accuracy Time Tradeoff Detailed Tables with Mean Values 128 E Progression for Last Two Submissions 142 F Enrollment Size 153 G Search Template Sizes 160 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING vii G.1 Mean Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 G.2 Median Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 H Template Creation Times 168 I Ranked Results 178 J Relative Combined Results 188 K Relative Accuracy and Number of Fingers 198 K.1 Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 K.2 Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 K.3 Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 K.4 All Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 L Combined Sorted Rankings for a Theoretical Use Case 207 List of Figures 1 NIST Biometric Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Example of Single-Finger Captures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 3 Example of an Identification Flat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 4 Examples of Live-Scan and Rescanned Ink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 NFIQ Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 6 Number of Submissions Received . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 7 Evaluation Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 9 Example of a Flipped Single-Index Finger Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 8 Example of a DET showing a Spike in FPIR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 10 Example of a Flipped Left Slap Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 11 Example of a Flipped Right Slap Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 12 Rank-sorted FNIR for Class A — Single Index Finger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 13 DET for Class A — Single Index Finger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 14 CMC for Class A — Single Index Finger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 15 Rank-sorted FNIR for Class A — Two Index Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 16 DET for Class A — Two Index Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 17 CMC for Class A — Two Index Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 18 Rank-sorted FNIR for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 19 DET for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 20 CMC for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 21 Rank-sorted FNIR for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 22 DET for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 23 CMC for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 24 FNIR and Search Time Scatter Plot — Class A — Left Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . . 43 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology viii FPVTE – F INGERPRINT M ATCHING 25 FNIR and Search Time Scatter Plot — Class A — Left Index — Greater Than or Equal To 20-Second Searches . . . . . . . 43 26 FNIR and Search Time Scatter Plot — Class A — Right Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . 44 27 FNIR and Search Time Scatter Plot — Class A — Right Index — Greater Than or Equal To 20-Second Searches . . . . . . 44 28 FNIR and Search Time Scatter Plot — Class A — Left and Right Index — Less Than 20-Second Searches . . . . . . . . . . 45 29 FNIR and Search Time Scatter Plot — Class A — Left and Right Index — Greater Than or Equal To 20-Second Searches . 45 30 Submission Round Progression — Class A — Left Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . . . . 46 31 Submission Round Progression — Class A — Left Index — Greater Than or Equal To 20-Second Searches . . . . . . . . . 46 32 Submission Round Progression — Class A — Right Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . . . 47 33 Submission Round Progression — Class A — Right Index — Greater Than or Equal To 20-Second Searches . . . . . . . . 47 34 Submission Round Progression — Class A — Left and Right Index — Less Than 20-Second Searches . . . . . . . . . . . . 48 35 Submission Round Progression — Class A — Left and Right Index — Greater Than or Equal To 20-Second Searches . . . 48 36 FNIR and Search Time Scatter Plot — Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 37 FNIR and Search Time Scatter Plot — Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 38 FNIR and Search Time Scatter Plot — Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 39 FNIR and Search Time Scatter Plot — Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 40 Submission Round Progression — Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 41 Submission Round Progression — Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 42 Submission Round Progression — Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 43 Submission Round Progression — Class B — IDFlat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 44 FNIR and Search Time Scatter Plot — Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 45 Submission Round Progression — Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 46 FNIR and RAM Used for Enrollment Set Tradeoff Scatter Plot for Class A — Left Index . . . . . . . . . . . . . . . . . . . 62 47 FNIR and RAM Used for Enrollment Set Tradeoff Scatter Plot for Class A — Right Index . . . . . . . . . . . . . . . . . . . 62 48 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class A — Left Index . . . . . . . . . . . . . . . . . . . . 63 49 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class A — Right Index . . . . . . . . . . . . . . . . . . . 63 50 FNIR and RAM Used for Enrollment Set Scatter Plot for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . 64 51 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class A — Left and Right Index . . . . . . . . . . . . . . 64 52 Enrollment Size in RAM Compared to On Disk for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . 65 53 Enrollment Size and Search Template Size in RAM Compared to On Disk for Class A — Left and Right Index . . . . . . . 65 54 FNIR and RAM Used for Enrollment Set for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 55 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class B — Identification Flats . . . . . . . . . . . . . . . 67 56 Enrollment Size in RAM Compared to On Disk for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . 68 57 Search Template Size in RAM Compared to On Disk for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . 68 58 FNIR and RAM Used for Enrollment Set Tradeoff Scatter Plot for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . 70 59 FNIR and RAM Used for Enrollment Set Tradeoff Scatter Plot for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . 70 60 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . 71 61 FNIR and Search Template Size in RAM Tradeoff Scatter Plot for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . 71 62 Enrollment Size and Search Template Size in RAM Compared to On Disk for Class C — Plain Impression . . . . . . . . . 72 63 Enrollment Size and Search Template Size in RAM Compared to On Disk for Class C — Plain Impression . . . . . . . . . 72 64 Enrollment Size and Search Template Size in RAM Compared to On Disk for Class C — Rolled Impression . . . . . . . . 73 65 Enrollment Size and Search Template Size in RAM Compared to On Disk for Class C — Rolled Impression . . . . . . . . 73 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ix 66 FNIR and Template Creation Time Tradeoff Scatter Plot for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . 75 67 FNIR and Template Creation Time Tradeoff Scatter Plot for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . 75 68 FNIR and Template Creation Time Tradeoff Scatter Plot for Class A — Left and Right Index . . . . . . . . . . . . . . . . . 76 69 FNIR and Template Creation Time Tradeoff Scatter Plot for Class B — Identification Flats . . . . . . . . . . . . . . . . . . 76 70 FNIR and Template Creation Time Tradeoff Scatter Plot for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . 77 71 FNIR and Template Creation Time Tradeoff Scatter Plot for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . 77 72 Rank-sorted FNIR for All Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 73 FpVTE 2003 SST Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 74 FpVTE 2003 MST Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 75 FpVTE 2003 LST Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 76 FpVTE 2003 LST Number of Fingers Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 77 DET for Participant C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 78 DET for Participant D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 79 DET for Participant E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 80 DET for Participant F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 81 DET for Participant G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 82 DET for Participant H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 83 DET for Participant I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 84 DET for Participant J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 85 DET for Participant K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 86 DET for Participant L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 87 DET for Participant M . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 88 DET for Participant O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 89 DET for Participant P . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 90 DET for Participant Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 91 DET for Participant S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 92 DET for Participant T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 93 DET for Participant U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 94 DET for Participant V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 95 DET for All Participants in All Classes — First Submissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 96 DET for All Participants in All Classes — Second Submissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 97 CMC for All Participants in All Classes — First Submissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 98 CMC for All Participants in All Classes — Second Submissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 99 Relative Combined Results — Left Index — Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 100 Relative Combined Results — Right Index — Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 101 Relative Combined Results — Left and Right Index — Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 102 Relative Combined Results — Left Slap — Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 103 Relative Combined Results — Right Slap — Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 104 Relative Combined Results — Left and Right Slap — Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 105 Relative Combined Results — Identification Flats — Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 106 Relative Combined Results — Ten-Finger Plain-to-Plain — Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology x FPVTE – F INGERPRINT M ATCHING 107 Relative Combined Results — Ten-Finger Rolled-to-Rolled — Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 108 Relative Accuracy — Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 109 Relative Accuracy — Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 110 Relative Accuracy — Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 111 Relative Accuracy — All Classes (Select Fingers) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 List of Tables 1 List of Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 3 Evaluation Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 4 Search and Enrollment Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 5 Number of Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 6 Finger Position Swapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 −3 — Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 FNIR @ FPIR = 10 −3 — Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 FNIR @ FPIR = 10 −3 — Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 FNIR @ FPIR = 10 −3 — Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 FNIR @ FPIR = 10 −3 — Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 FNIR @ FPIR = 10 −3 — Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 13 FNIR @ FPIR = 10 −3 — Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 14 FNIR @ FPIR = 10−3 — Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 7 FNIR @ FPIR = 10 8 9 10 11 12 FNIR @ FPIR = 10 −3 — Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 16 FNIR @ FPIR = 10 −3 — Class C — Ten-Finger Plain-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 17 Median Identification Times for Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 18 Median Identification Times for Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 19 Median Identification Times for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 20 Median Identification Times for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 21 Ranked Results for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 22 Ranked Results for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 23 Ranked Results for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 24 Median Identification Time Results for Class A — Left Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . 115 25 Median Identification Time Results for Class A — Left Index — Greater Than or Equal To 20-Second Searches . . . . . . 116 26 Median Identification Time Results for Class A — Right Index — Less Than 20-Second Searches . . . . . . . . . . . . . . 117 27 Median Identification Time Results for Class A — Right Index — Greater Than or Equal To 20-Second Searches . . . . . . 118 28 Median Identification Time Results for Class A — Left and Right Index — Less Than 20-Second Searches . . . . . . . . . 119 29 Median Identification Time Results for Class A — Left and Right Index — Greater Than or Equal To 20-Second Searches 30 Median Identification Time Results for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 31 Median Identification Time Results for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 32 Median Identification Time Results for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 33 Median Identification Time Results for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 15 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 120 FPVTE – F INGERPRINT M ATCHING xi 34 Median Identification Time Results for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . 125 35 Median Identification Time Results for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . 126 36 Median Identification Time Results for Class C — Ten-Finger Plain-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . 127 37 Mean Identification Time Results for Class A — Left Index — Less Than 20-Second Searches . . . . . . . . . . . . . . . . 129 38 Mean Identification Time Results for Class A — Left Index — Greater Than or Equal To 20-Second Searches . . . . . . . . 130 39 Mean Identification Time Results for Class A — Right Index — Less Than 20-Second Searches 40 Mean Identification Time Results for Class A — Right Index — Greater Than or Equal To 20-Second Searches . . . . . . . 132 41 Mean Identification Time Results for Class A — Left and Right Index — Less Than 20-Second Searches . . . . . . . . . . 133 42 Mean Identification Time Results for Class A — Left and Right Index — Greater Than or Equal To 20-Second Searches . . 134 43 Mean Identification Time Results for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 44 Mean Identification Time Results for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 45 Mean Identification Time Results for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 46 Mean Identification Time Results for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 47 Mean Identification Time Results for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 48 Mean Identification Time Results for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . 140 49 Mean Identification Time Results for Class C — Ten-Finger Plain-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . 141 50 Progression of Identification Timing and Accuracy for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . 143 51 Progression of Identification Timing and Accuracy for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . 144 52 Progression of Identification Timing and Accuracy for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . 145 53 Progression of Identification Timing and Accuracy for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 54 Progression of Identification Timing and Accuracy for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . 147 55 Progression of Identification Timing and Accuracy for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . 148 56 Progression of Identification Timing and Accuracy for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . 149 57 Progression of Identification Timing and Accuracy for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . 150 58 Progression of Identification Timing and Accuracy for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . 151 59 Progression of Identification Timing and Accuracy for Class C — Ten-Finger Plain-to-Rolled . . . . . . . . . . . . . . . . 152 60 Storage and RAM Requirements for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 61 Storage and RAM Requirements for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 62 Storage and RAM Requirements for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 63 Storage and RAM Requirements for Class B — Identification Flats 64 Storage and RAM Requirements for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 65 Storage and RAM Requirements for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . 159 66 Mean Search Template Sizes for Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 67 Mean Search Template Sizes for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 68 Mean Search Template Sizes for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 69 Median Search Template Sizes for Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 70 Median Search Template Sizes for Class B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 71 Median Search Template Sizes for Class C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 72 Enrollment Time Results for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 73 Enrollment Time Results for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 74 Enrollment Time Results for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 . . . . . . . . . . . . . . . 131 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology xii FPVTE – F INGERPRINT M ATCHING 75 Enrollment Time Results for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 76 Enrollment Time Results for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 77 Enrollment Time Results for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 78 Enrollment Time Results for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 79 Enrollment Time Results for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 80 Enrollment Time Results for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 81 Ranked Results for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 82 Ranked Results for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 83 Ranked Results for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 84 Ranked Results for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 85 Ranked Results for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 86 Ranked Results for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 87 Ranked Results for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 88 Ranked Results for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 89 Ranked Results for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 90 Operational Ranking for Class A — Left Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 91 Operational Ranking for Class A — Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 92 Operational Ranking for Class A — Left and Right Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 93 Operational Ranking for Class B — Left Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 94 Operational Ranking for Class B — Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 95 Operational Ranking for Class B — Left and Right Slap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 96 Operational Ranking for Class B — Identification Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 97 Operational Ranking for Class C — Ten-Finger Plain-to-Plain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 98 Operational Ranking for Class C — Ten-Finger Rolled-to-Rolled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING xiii Executive Summary FpVTE was conducted primarily to assess the current capabilities of fingerprint matching algorithms using operational datasets containing several million subjects. There were three classes of participation that examined one-to-many identification using various finger combinations from single finger up to ten fingers. Class A used single-index finger capture data and evaluated single index finger (right or left) and two index finger (right and left) identification. Class B used identification flat (IDFlat) captures (4-4-2; left slap, right slap, and two thumbs simultaneously) and evaluated ten-finger, eight-finger (right and left slap), and four-finger (right or left slap) identification. Class C used rolled and plain impression (4-4-1-1; left slap, right slap, left thumb, and right thumb) captures and evaluated ten-finger rolled-to-rolled, ten-finger plain-to-plain, and ten-finger plain-to-rolled identification. Enrollment sets used for one-to-many identification varied in size from 5 000 up to 5 000 000 enrolled subjects. Any segmentation of four-finger slap images or two-thumb captures was performed by the submitted software. All data used was sequestered operational data that was not shared with any of the participants. The evaluation allowed each participant to make two submissions per class (A, B, and C) of participation over three rounds. After each of the first two rounds of submissions, feedback was provided to the participants and they were allowed to evaluate their performance, make adjustments to their submissions, and resubmit for the next round. The results of the third and final round of submissions are reported in this document. The evaluation was conducted at the National Institute of Standards and Technology (NIST) using commodity NISTowned hardware. Participant submissions were compliant to the testing Application Programming Interface (API), which were linked to a NIST-developed test driver and run by NIST employees. All submissions went through validation testing to ensure that results generated on NIST’s hardware matched results participants generated on their own hardware. This was the first large-scale one-to-many fingerprint evaluation since FpVTE 2003. In 2003, participants brought their own hardware to NIST to process the evaluation data. The datasets in 2003 had approximately 25 000 subjects and required millions of single subject-to-subject matches. The current FpVTE used a testing model closer to real one-to-many identification systems by allowing the submitted software to control how it does the one-to-many search and return a candidate list of potential matches. The number of subjects used was also significantly higher, as the current FpVTE had ≈ 10 million subjects in the testing datasets. The results in this report are based on 30 000 (10 000 mates and 20 000 nonmates) search subjects. There will be an additional report with results (lower errors rates) using 350 000 (50 000 mates and 300 000 nonmates) search subjects. In addition to measuring current performance capabilities of one-to-many identification algorithms, FpVTE was conducted to: . study open-set identification versus enrolled sample sizes extending into the multiple millions; . provide a testing framework and API for enrollment sizes that must be spread across the memory of multiple compute nodes; . evaluate on operational datasets containing newer data from live-scan ten-finger “identification flat” capture systems, other live-scan capture devices (e.g., single-finger and multi-finger), and historically significant scanned inked fingerprints; . analyze one-to-many identification accuracy, speed, template size, number of fingers, enrollment set sizes, and computational resources; . create a fingerprint testing data repository with a vast majority of data errors corrected. FpVTE was not intended to: . measure performance of an operational Automated Fingerprint Identification System (AFIS); . evaluate scanners or acquisition devices; . evaluate latent fingerprint or mobile-captured data. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology xiv FPVTE – F INGERPRINT M ATCHING Key results from FpVTE were: . Fingerprint Identification Accuracy: The most accurate fingerprint identification submissions achieved false negative identification rates (FNIR, or “miss rates”) of 1.9% for single index fingers, 0.27% for two index fingers, 0.45% for four-finger identification flats (IDFlats), 0.15% for eight-finger IDFlats, 0.09% for ten-finger IDFlats, 0.1% for tenfinger rolled-to-rolled, 0.13% for ten-finger plain-to-plain, and 0.11% for ten-finger plain-to-rolled. These numbers are reported at a false positive identification rate (FPIR) of 10−3 . 30 000 search subjects were used for these results (10 000 mates and 20 000 nonmates). The number of enrolled subjects used for single index fingers was 100 000, 1.6 million for two index fingers, 3 million for IDFlats, and 5 million for ten finger plains and rolled. A larger search set (50 000 mates and 300 000 nonmates) is being completed so that an even smaller FPIR can be reported. Those results, when available, will be included in an additional report. Section 7. . Accuracy versus Speed: The fastest submissions were not the most accurate. The most accurate submissions showed the ability to decrease search times with minimal loss in accuracy. The format of this evaluation may not have fully explored the lower limit of search speed with minimal loss in accuracy. Section 8 and Appendix E. . Number of Fingers: Not surprisingly, using more fingers improved accuracy. In fact, the most accurate results were achieved with ten fingers, searching against the largest enrollment sets of 3 and 5 million subjects. An interesting result that needs further study and analysis is that two-index finger accuracy was better than four-finger IDFlats. Based on some analysis of missed mates, it appears image quality may have played some roll in this result. Section 11. . Computation Resources: The most accurate submissions were able to achieve their results with similar Random Access Memory (RAM) usage to other submissions. These same accurate submissions typically took longer to enroll (i.e., extract features) the fingerprint images used in the evaluation. Section 9. . Candidate Lists: For most top performers, a majority of the time, the mate appeared within the top three candidates of the candidate list or did not appear at all. Section 7. . Ranked Results: Results that group FNIR, enrollment time, search time, and RAM usage all in a single table are shown in Section 10 and Appendix I. The tables are rank-sorted on FNIR, but include ranks in all the other categories for cross-category comparison. . Number of Subjects in Enrollment Set: Results for different enrollment set population sizes will not be available until the larger search sets are complete, but the initial results across classes of participation showed that eight- and ten-finger accuracy was superior versus larger enrollment sets. In fact, the most accurate identification results were always achieved with the ten-finger search sets. Section 7. . Data Type: There were three classes of participation in which data from single fingers, IDFlats, and “legacy” tenfinger rolled and plain impression data types were evaluated. Results of IDFlats and ten-finger rolled and plain showed little variation in the accuracy of the top-performing submissions. This means the best performers could tune their submissions to accurately match all the data types evaluated. Section 10. . Accuracy Gap: The “gap” between the most accurate submissions and the “next tier” appears to be much closer than in FpVTE 2003. Sections 10 and 12. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING xv Caveats 1. Specific nature of the biometric data: The absolute error rates quoted here were measured over a very large fixed corpus of operational fingerprint images. The error rates measured here are realistic if the submissions are applied to this kind of data. However, in other applications, the applicability of the results may differ due to a number of factors legitimately not reflected in the FpVTE experimental design. Among these are: . how much slap fingerprint segmentation errors contribute to core matching accuracy; . algorithmic limitations caused by the FpVTE API; . unknown bugs in the submission; . image quality from using a different data source. 2. Not an Automated Fingerprint Identification System (AFIS) test: While this evaluation is intended to measure the core capabilities of matching algorithms in a large one-to-many scenario, it is not intended to be a full assessment of an operational AFIS. 3. Timing of the submissions: While every attempt was made to perform timing on the exact same hardware with the exact same conditions, it is possible that certain functions of the operating system could have had an unintended negative effect on timing results. The timing operations reported for each submission submitted were performed in the exact same manner. Generous “cutoff” times were employed to prevent wasting compute cycles on submissions whose API functions never returned a value. 4. Aggregate finger positions: When reporting estimated template RAM usage statistics, participants were permitted to return an aggregate template size when the input image contained more than one finger (e.g., “slaps”). Any statistics over these types of images were reported as-is and the values were not divided by the number of fingers expected for the capture type. While these participants are not denoted in the report, the actual RAM usage may be a better statistic for comparison. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology xvi FPVTE – F INGERPRINT M ATCHING Release Notes All FpVTE related reports, drafts, announcements, and news items may be found on the website http://fingerprint.nist.gov. . Application Program Interface and Test Plan: The FpVTE API [15] contains additional details about creating a submission compatible with the FpVTE test driver. All submissions tested in FpVTE from 2012-2014 were fully compatible with the FpVTE API and linked without modification to the FpVTE test driver. . Appendices: Appendix A has full-scale plots for individual participant results that some may prefer to the grouped plots in the main body of the report. Appendices B through L have complete sets of tables for various analyses to help reduce the number of tables in the main body of the report. . Submission identifiers: Throughout this report, submissions are identified by letter code. For reference, the letters are associated with the providers’ names in a running footnote. . Typesetting and Graphics: Virtually all of the content in this report was produced automatically. This involved the R graphs directly from the FpVTE test driver’s output. Other use of scripting tools to generate LATEX content and graphics were produced with TikZ. Use of these technologies improved timeliness, flexibility, maintainability, and reduced transcription errors. . Evaluation Data Ground Truth: Unknown mates within and across datasets create a significant problem and time delay for large one-to-many data testing. . Contact: Correspondence regarding this report should be directed to FPVTE at NIST dot GOV. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 1 1 Introduction In 2012, NIST launched a new Fingerprint Vendor Technology Evaluation (FpVTE) with two main goals. The first goal was to assess the current capabilities of matching algorithms using operational datasets with several million subjects. The second goal was to evaluate different operational considerations that could impact matching accuracy. These considerations included number of fingers used, data types (live-scan, single-finger capture, slap capture requiring segmentation, and rolled), number of enrolled subjects, and matching speeds. Evaluating biometric capabilities is an important task, particularly for fingerprint identification, given its widespread applications. Large-scale evaluations of core accuracy and functionality of biometric recognition algorithms using operational data will not only reveal the capabilities of the current state of the art but can also identify the limitations and gaps of the current algorithms. The former sets realistic operational expectations and the latter directs future research to improve and enhance current technologies. FpVTE was conducted by NIST and sponsored by the Department of Homeland Security (DHS) and the Federal Bureau of Investigation (FBI). All work was performed at the NIST Gaithersburg facility using hardware owned by NIST. FpVTE includes three rounds of submissions, with participants making algorithm adjustments based on performance reports after the first two submissions. Participant submissions were required to conform to the test plan API [15]. The evaluation had three classes of participation. Class A used single-index finger capture data and evaluated single index fingers searched against 5 000 up to 100 000 enrolled subjects (“single-finger identification”), and two index fingers searched against 10 000 up to 1 600 000 enrolled subjects (“two-finger identification”). Class B used IDFlat captures (44-2) and evaluated ten-finger, eight-finger (right and left slap), and four-finger (right or left slap) identification searched against 500 000 up to 3 000 000 enrolled subjects. Class C used rolled and plain impression (4-4-1-1) captures and evaluated ten-finger rolled impression, ten-finger plain impression, and ten-finger plain impression matched to rolled impression identification searched against 500 000 up to 5 000 000 enrolled subjects. Any segmentation of four-finger slap images or two-thumb captures was performed by the submission. Biometric identification is defined as, “[the] process of searching against a biometric enrollment database to find and return the biometric reference identifier(s) attributable to a single individual” [7]. Biometric identification is broadly categorized into closed-set and open-set identification. Closed-set identification refers to cases where all searches have a corresponding enrolled mate in the biometric enrollment database. An example of a closed-set identification application is a cruise ship on which all passengers are enrolled. The outcome of a closed-set identification subsystem is a candidate list that contains the identity of one or more enrolled individuals whose enrolled samples are most similar to the search (query) sample. Ideally, the correct mate appears in the first rank. As such, the primary accuracy metric for closed-set identification is hit rate (or its complement, miss rate = 1.0 − hit rate), which is the fraction of times the system returns the correct identity within the specified top ranks. In open-set identification, not all searches have a corresponding enrolled mate in the biometric enrollment database [3]. The expected outcome of an open-set identification subsystem is a candidate list of L closest (or most similar) enrolled identities when the search sample is from an enrolled individual, or an indication that the search sample is from an individual not in the biometric enrollment database. Therefore, primary accuracy metrics for an open-set identification are false positive identification (false alarm or Type I error) rate and false negative identification (miss or Type II error) rate. These metrics are described in Section 6. Closed-set identification applications are very limited because in the majority of real-world identification applications, not all individuals are or can be enrolled. Most real-world biometric identification applications, such as searches against a watch-list or searches for first-time arrestees, are open-set identification. For that reason, FpVTE only evaluated open-set identification algorithms. This document reviews metrics for evaluating the performance of open-set identification algorithms and reports performance for submissions from 18 participants. Three participated in Class A only and the other 15 submitted for all three classes. Four participants withdrew before making a submission for testing. Table 1 shows the list of participants and in which classes they made submissions for evaluation. These participants are referenced at the bottom of every page with their assigned identification letter. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 2 FPVTE – F INGERPRINT M ATCHING ID Name Participation Class C D E F G H I J K L M O P Q S T U V afis team 3M Cogent Neurotechnology Papillon Dermalog Hisign Bio-Info Institute NEC Sonda Tiger IT Innovatrics SPEX ID Solutions id3 Morpho Decatur Industries BIO-key Aware AA Technology A, B, C A, B, C A, B, C A, B, C A, B, C A, B, C A, B, C A, B, C A A, B, C A, B, C A, B, C A A, B, C A, B, C A A, B, C A, B, C Table 1: Participant IDs, names, and classes of submission for evaluation. Section 2 has some history on fingerprint evaluations performed at NIST. Section 3 describes the data used in FpVTE. Section 4 describes the protocol used for submission acceptance and testing. Section 5 gives details on the two-stage matching approach used in FpVTE. Sections 6 through 11 talk about the metrics used to measure accuracy and report the results from testing. While recognition error rates are important and widely reported, computational resources required by algorithms are a significant aspect of performance, especially for large-scale operations. To that end, we report the computation time, storage requirements, and their accuracy tradeoffs for each of the submissions. Finally, Section 12 examines some FpVTE 2003 results, then Section 13 and Section 14 talk about some lessons learned and future plans. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 2 2.1 3 History and Motivation NIST Biometric Evaluations The first one-to-many fingerprint evaluation conducted at NIST was FpVTE 2003. This evaluation required that participants bring their own hardware and software to NIST for the evaluation. NIST supplied the data and retained all matching results for final analysis. FpVTE 2003 had three classes of participation: Small-Scale, Medium-Scale, and LargeScale. Small-Scale testing used 1 000 single-finger capture images resulting in 1 million subject-to-subject comparisons. Medium-Scale testing used 10 000 single-finger captures resulting in 100 million comparisons. Large-Scale testing used 25 000 subjects and various combinations of fingers resulting in 1.044 billion comparisons. The evaluation had 18 participants that submitted 34 systems for testing. NIST has conducted several fingerprint-related evaluations in the last decade (Figure 1). The first fingerprint evaluation was called Proprietary Fingerprint Template 2003 (PFT 2003). PFT 2003 was a one-to-one matching evaluation that looked at the core matching capabilities of fingerprint matching software. It did not evaluate one-to-many capabilities. In 2010, NIST changed the name of PFT 2003 to PFTII, utilizing newer, larger datasets and reporting information on timing and template sizes, in addition to accuracy. Minutiae Exchange (MINEX), also a one-to-one matching evaluation, began in 2005. MINEX was started to support testing of fingerprint matching technologies using INCITS 378 standard interoperable templates [6]. About a year later, Ongoing MINEX was created to support Personal Identity Verification (PIV) by establishing guidelines and measuring accuracy for interoperable template encoders and matchers. In addition to fingerprint-related evaluations, NIST has performed evaluations for other biometrics such as face and iris. Additional information and links for all the NIST biometric related evaluations can be found on the NIST biometric evaluations website [5]. FRVT SlapSeg FRVT Biometric Usability ICE 2002 2004 2006 IREX I IQCE MBE PFTII FpVTE FRVT IREX IV FIVE Tatt-C 2008 2010 2012 2014 2003 2005 2007 2009 2011 FpVTE PFT MINEX FRGC ICE ELFT MINEX II MBGC SlapSeg II ELFT MBE IREX III NFIQ2 2013 Figure 1: NIST biometric evaluations. 2.2 Purpose One of the main purposes of this FpVTE evaluation was to provide a refresh on the testing performed in 2003 and allow an opportunity for participation by organizations that missed the previous evaluation. There had been many inquiries in the past several years on when NIST would perform a similar evaluation. Additionally, the dataset size for FpVTE 2003 was around 25 000 total subjects. The current FpVTE testing used enrollment sets ranging from 10 000 subjects to 5 million subjects. NIST has already conducted one-to-many biometric evaluations with enrollment set sizes over 1 million subjects for other modalities (e.g., Face Recognition Vendor Test (FRVT)/Multiple Biometric Evaluation (MBE) for face and Iris Exchange Evaluation (IREX III) for iris), in which the sizes of the enrollment templates allowed the entire enrollment set to fit into the RAM of a single compute node. FpVTE was the first biometric evaluation at NIST that added the ability to partition the enrollment set across multiple compute nodes, expanding the possibilities in size and breadth of enrollment sets. In addition to broadening the enrollment set, FpVTE strived to: C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 4 FPVTE – F INGERPRINT M ATCHING . assess the current performance of one-to-many fingerprint matching software using operational fingerprint data; . provide a testing framework and API for enrollment sizes that must be spread across the memory of multiple compute nodes; . support US Government and other sponsors in setting operational thresholds; . evaluate on operational datasets containing newer data from live-scan ten-finger IDFlat capture systems, other livescan capture devices (e.g., single-finger and multi-finger), and historically significant scanned inked fingerprints; . analyze one-to-many identification accuracy versus, speed, template size, number of fingers, enrollment set sizes, and computational resources. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 3 3.1 5 Data Classes of Participation FpVTE was separated into three classes of participation: A, B, and C. All participants were required to make a Class A submission. Along with the Class A submission, a participant could additionally participate in Class B, or both Class B and Class C. These were the only three participation combinations available. Images were captured via live-scan sensor and rescanned ink. A live-scan sensor refers to the type of sensor that digitally records the friction ridges of a finger through techniques such as electrical or optical sensing. Scanned ink is the process of creating a digital image by using an image scanner to optically capture from paper images of friction ridges created by a finger covered with ink. Class A consisted of live-scan single-finger captures of the left and right index fingers (Figure 2). Figure 2: Example of single-finger captures of left and right index fingers. Class B consisted of live-scan IDFlats, which captured left and right four-finger slaps and simultaneous left and right thumbs, known as “4-4-2” (Figure 3). Figure 3: Example of an identification flat (4-4-2) capture of a left and right slap and left and right thumbs. Class C consisted of the rolled and plain impressions from a more traditional 14 print card/record and was a mix of live-scan and rescanned ink (Figure 4). Figure 4: Example of live-scan and rescanned ink 14 print card/record that include rolled and plain impressions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 6 3.2 FPVTE – F INGERPRINT M ATCHING Datasets The evaluation datasets used in FpVTE were from anonymized operational datasets and made available to NIST for fingerprint evaluations. The datasets are for government use only and will not be released to the public. The datasets will, to the extent permitted by law, be protected under the Freedom of Information Act (5 U.S.C. § 522) and the Privacy Act (5 U.S.C. § 522a) as applicable. The datasets were comprised of several fingerprint impression types, including rolled, multi-finger plains, and singlefinger plains. Rolled images were all individual captures that attempted to record the full width of the fingerprint by rolling from side-to-side during capture. Multi-finger plains captured the four right and four left fingers at the same time. For identification flats, the two thumbs were captured at the same time. Single-finger plains were individual captures of the right and left index fingers on a single-finger capture device. Many of the datasets were larger samples of data used in previous NIST evaluations, such as PFT, MINEX, NIST Fingerprint Image Quality (NFIQ), and FpVTE 2003. The single-finger capture and identification flat fingerprint images were provided by DHS. The ten-finger rolled and slap fingerprint images included data from the FBI, DHS, Los Angeles County Sheriff’s Department (LACNTY), Arizona Department of Public Safety (AZDPS), and Texas Department of Public Safety (TXDPS). Table 2 shows the source of data for each class in the evaluation. Class Dataset Enroll Mate Enroll Nonmate Search Mate Search Nonmate A VISIT-I/POEBVA DHS2 VISIT-II 25% 25% 50% 5% 5% 90% 25% 25% 50% 25% 25% 50% B VISIT-II PDR_IDF 96% 4% 85% 15% 96% 4% 62.5% 37.5% C AZDPS INSBEN LACNTY TXDPS PDR-Roll 33.3% 0% 33.3% 0% 33.3% 4% 11.5% 30% 11.5% 43% 33.3% 0% 33.3% 0% 33.3% 18.75% 12.5% 18.75% 12.5% 37.5% Table 2: Percentage of data used from each source. All images in the datasets were 8-bit grayscale. Images were previously compressed using Wavelet Scalar Quantization (WSQ) compression [2], but were passed to the submitted software as reconstructed raw pixel images, decompressed using libwsq from NIST’s NIST Biometric Image Software (NBIS) distribution [14]. All images were scanned at 500 pixels per inch. The dimensions of the images varied, but were provided as input information to the participant’s submission. The distribution of NIST Fingerprint Image Quality (NFIQ) algorithm [12, 13] values for the evaluation datasets is shown in Figure 5. NIST Fingerprint Segmentation algorithm (NFSEG) was used to segment slap impression images into individual fingers before computing NFIQ values. Multiple-finger plain captures were not segmented. Submissions were required to perform segmentation of fingerprints, if necessary. Subjects with missing fingers were not removed from the dataset. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Class Fingers Type 1 2 3 4 5 A L/R Index L/R Index Enroll Search 42.8% 39.5% 32.5% 32.9% 20.2% 20.7% 2.2% 2.2% 2.3% 4.7% B Left Slap Right Slap L/R Slap IDFlats Search Search Search Search 44.9% 50.2% 47.6% 48.3% 26.8% 24.4% 25.6% 25.7% 16.0% 13.0% 14.5% 14.2% 6.9% 5.8% 6.3% 6.4% 5.4% 6.6% 6.0% 5.4% Ten Plain Ten Rolled Ten Rolled Search Enroll Search 47.8% 34.2% 35.4% 26.6% 19.5% 21.3% 16.2% 25.0% 23.0% 5.5% 8.6% 9.1% 3.9% 12.7% 11.2% C 7 100% 75% 50% 25% 0% L/R Index Left Slap Right Slap L/R Slap ID Flats Ten Plain Ten Rolled Finger NFIQ 1 2 3 4 5 Figure 5: NFIQ distribution for datasets, after segmentation, where 1 is highest quality and 5 is lowest quality. 3.3 Evaluation Scenarios The three classes of participation had various data type and fingerprint combinations that could be evaluated as summarized in autoreftab:data-scenarios. The contents of Table 4 shows details of the search and enrollment sets used during the evaluation for each class of participation. The final column of Table 4 shows the various sizes of the datasets that were tested. The subjects reserved for the search sets contained 200 000 with a known mate and 400 000 with no mate in the enrollment set. This report is based on a random sample of 10 000 mated and 20 000 nonmated searches from the full search set. Class S1 x E1 A S2 x E2 S3 x E3 S4 x E4 B Table 3 shows the various combinations in which the search sets were searched against the enrollment sets listed in Table 4. Those combinations are also described for each class as follows. S5 x E4 S6 x E4 S7 x E4 S8 x E5 C . Class A — Index Fingers Scenario S9 x E5 S9 x E6 – One plain index finger searched against an enrollment set of one plain index Table 3: Evaluation scenarios. fingers. The plain images were from single-finger captures of left and right Refer to Table 4 for descriptions index fingers. of the search and enrollment set – Two plain index fingers searched against an enrollment set of two plain index codes. fingers. The plain images were from single-finger captures of left and right index fingers. . Class B — Identification Flats – Four-, eight-, and ten-finger identification flats (4-4-2, left/right four-finger plain impressions, and a two-thumb plain impression) searched against an enrollment set of ten-finger identification flats. Any segmentation was performed by the submission. . Class C — Ten-Finger Rolled/Slap – Ten-finger rolled impressions searched against an enrollment set of ten-finger rolled impressions. – Ten-finger plain impressions searched against an enrollment set of ten-finger plain impressions. Plain impression images were 4-4-1-1 (left/right four-finger plain impression and left/right single-thumb plain impressions). Any segmentation was performed by the submission. – Ten-finger plain impressions searched against an enrollment set of ten-finger rolled impressions. Plain impression images were 4-4-1-1 (left/right four-finger plain impressions and left/right single-thumb plain impressions). Any segmentation was performed by the submission. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 8 FPVTE – F INGERPRINT M ATCHING Class Search A Set Description # Images # Fingers Impression S1 Right Index 1 1 Plain S2 Left Index 1 1 Plain S3 Left and Right Index 2 2 Plain S4 Right Slap IDFlat 1 4 Plain S5 Left Slap IDFlat 1 4 Plain S6 Left and Right Slap IDFlat 2 S7 Identification Flats 3 S8 Ten Finger Rolled 10 S9 Ten Finger Plain 4 E1 Right Index 1 1 Plain E2 Left Index 1 1 Plain E3 Left and Right Index 2 2 Plain E4 Identification Flats 3 E5 Ten Finger Rolled 10 E6 Ten Finger Plain 4 B C Enrollment A B 8 (4, 4) 10 (4, 4, 2) 10 10 (4, 4, 1, 1) 10 (4, 4, 2) 10 Plain Plain Rolled Plain Plain Rolled C 10 (4, 4, 1, 1) Plain # Subjects Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 Mate: 10 000 Nonmate: 20 000 10 000 100 000 10 000 100 000 100 000 500 000 1 600 000 500 000 1 600 000 3 600 000 500 000 1 600 000 3 000 000 5 000 000 500 000 1 600 000 3 000 000 5 000 000 Table 4: Search and enrollment datasets used in FpVTE. Class refers to the class of participation, as defined in Subsection 3.1. Set is an identifier used to uniquely identify the various FpVTE datasets in a concise manner. Description indicates the finger combinations composing each dataset. # Images specifies the number of images per subject. # Fingers shows the maximum number of fingers per subject. If a subject’s fingers were spread across multiple images, the maximum number of fingers for each image is also shown. Impression refers to the impression type of the imagery in each dataset. Impressions could be either plain or rolled, as defined in Subsection 3.2. # Subjects shows the number of subjects in each dataset. For ’Search’ datasets, the value has been split into mate and nonmate subjects, referring to whether or not a known mate for the subject exists in the corresponding enrollment set. For ’Enrollment’ datasets, the # Subjects show the various enrollment set sizes planned for evaluation. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 3.4 9 “Size” of the Test The overall size of the test had implications on total run time. Table 5 shows the number of enrollments and comparisons that were completed for each submission. Time limits allowed for a maximum of 3 seconds per finger for each enrollment, and up to 500 seconds for Class A and 90 seconds for Class B and C to complete each search. Participants could make two submissions for each class of participation. Three of the 18 participants submitted for Class A only, while the other 15 submitted for all three classes, creating a total of 96 submissions analyzed in this report. There were three rounds of submissions, with participants receiving results after the first two submissions to analyze before making their next submission. Most submissions shared enrollment templates between both submissions within a class, but two or three submissions in each round required generating enrollment templates separately. Class Enrollments Performed Subject to Subject Comparisons A 5.8 54 B 45.5 362 C 113 452 Table 5: Approximate number (in millions) of enrollments and comparisons that were completed to evaluate each submission for that class of participation. One unexpected event was that the majority of the submissions required re-enrolling the data for each round of submissions. Each round of submissions required approximately 2.5 billion fingerprint enrollments. It took two to three months for each round of submissions to enroll all the data for all the submissions before searching could begin. The 30 000 (20 000 nonmate and 10 000 mate) searches took another one to two months to complete for each round of submissions for all classes across all submissions and finger combinations. The first two rounds resulted in approximately 43 trillion subjectto-subject comparisons, and the final round will have approximately 470 trillion subject-to-subject comparisons. The final submission will be the only one to do searching against variable enrollment set sizes. Total enrollment and search times were slightly longer when accounting for submissions that failed in the middle of enrollment or matching and had to be fixed and restarted (Subsection 4.4). C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 10 4 4.1 FPVTE – F INGERPRINT M ATCHING Experiment and Test Protocol API Overview The FpVTE API consisted of three major steps: feature extraction, finalization, and identification. During all steps of the evaluation, participants had read-only access to a configuration directory, where configuration files, models, or other algorithm-specific data could be stored. Not all participants made use of the configuration directory. It was made available as a convenience to those participants who did not wish to store customized traits about their software inside the submission itself. In practice, many participants were able to perform changes, such as adjusting identification speed, simply by changing a configuration file within this directory instead of needing to recompile their submission. 4.1.1 Feature Extraction The first step in the FpVTE evaluation pipeline was feature extraction. During this step, software submissions were given the opportunity to turn one or more images of fingerprints into a single fingerprint template. There were no requirements on the format of the fingerprint template, as it was expected that the majority of participants would make use of proprietary template formats. The FpVTE API provided a way to distinguish between the images that would be used for searching and the images used to compose the enrollment set, though participants were free to treat all images in the same fashion. For each instance of feature extraction, the FpVTE test driver called an initialization method a single time, giving the submission an opportunity to load information that might be needed for feature extraction. After the initialization call, the feature extraction method was called N times. The FpVTE API provided the finger position, impression type, NFIQ value (no value provided for slap images), image dimensions, and raw image bytes for each input image during the call to extract features. As output, participants were to return the finger position, an image quality value, a fingerprint template, and the size of the fingerprint template as it would be stored in RAM. The RAM size was important to determine the number of compute nodes needed during identification, because the size of the template on disk might be significantly larger or smaller than the size of the template in RAM. Participants could optionally return a core and delta coordinate for the image. 4.1.2 Finalization As fingerprint templates were returned from feature extraction, the FpVTE test driver added them to a RecordStore (a key-value pair storage mechanism of the Biometric Evaluation Framework). This allowed I/O operations to be excluded when calculating the runtime of feature extraction, as well as to allow NIST to store the large quantity of template data in the most efficient way on NIST’s hardware. Once all features were extracted from enrollment set fingerprint images, the FpVTE test driver called an API method to “finalize” the enrollment set. Based on the sum of the sizes of templates in RAM, NIST calculated the appropriate number of compute nodes and passed this information, along with the RecordStore of all templates, to the finalization method. During this method, the submission was to divide the enrollment set into a partition for each compute node, as well as perform any sort of indexing, statistics, or other pre-identification tasks required. Unlike how NIST controlled the storage mechanism for templates in feature extraction, participants had full control of how to store their finalized enrollment sets. In many cases, the size of the finalized enrollment set was either much larger or much smaller than the sum of the fingerprint templates sizes returned from feature extraction as stored on disk. 4.1.3 Identification Once the finalized set was created, its read-only root directory was provided to the identification methods. Searching the finalized set for candidates occurred over two stages. The first stage was performed on separate compute nodes (if necessary, as determined by the amount of RAM needed), potentially searching subsets of the enrollment set. Output from C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 11 this first stage was coalesced and provided to the second stage, which took place on a single compute node. These two stages of identification are described at length in Section 5. 4.2 Test Constraints In order to complete the evaluation in a reasonable amount of time, as well as to mimic potential operational requirements, certain constraints were placed on FpVTE submissions. All times are maximum averages over the pre-evaluation datasets (Section 3), though NIST employed reasonable “cutoff” times to prevent processes from never completing. See Caveats for additional information about timing. . Operating System and Compilation Environment: All submitted implementations required 64-bit linkage and were tested on CentOS 6.2 (kernel version 2.6.32-220.7.1.el6.x86_64). NIST linked participant submissions to the FpVTE test driver written in C++ with GNU g++ 4.4.6-3, using glibcxx 3.4.13 and glibc 2.12. . Evaluation Hardware: Timing computations were performed on a Dell M610 with two Intel X5690 3.47 GHz processors and 192 GB of RAM. . Threaded Computations: Threaded computations were only allowed for the finalization step. All other functions were not to perform multithreaded computations, as the FpVTE test driver handled parallelism efficiently on the NIST compute nodes. . Feature Extraction Time: Feature extraction was required to complete in 3 seconds or less for each input fingerprint. A four-finger slap was counted as four input fingerprints. . Finalization Time: Finalization of the enrollment set of fingerprint templates was required to complete in 12 hours or less on a single compute node. . Search Time: Search time is the combined time measurement for stage one and stage two identification (not including initialization times). For Class A data, searches needed to complete in under 500 seconds, though implementations that searched in under 20 seconds were reported separately. For Class B and Class C, all searches needed to complete in under 90 seconds. 4.3 Biometric Evaluation Framework The FpVTE test driver made use of several C++ classes that are part of the NIST Image Group’s Biometric Evaluation Framework [11] designed to make writing code for running biometric evaluations easier and more efficient, especially on NIST-owned equipment. Classes from the framework used in the FpVTE API included key-value pair file storage, safe dynamic arrays, error handling, and more. While the framework was mainly provided to participants as a means of interacting with the FpVTE API, many participants chose to use some classes internally in their submissions. In order to process the massive amount of data required for the evaluation, the submission was executed as a scalable parallel job. Within the NIST testbed, an implementation of the Message Passing Interface (MPI) [10] was used to execute the evaluation test programs. By using the MPI software, the size of the parallel job (in terms of participating compute nodes) can be matched to the size of the input dataset. The framework supports parallelism by abstracting and hiding the lower-level communication, error handling, and other facets of the MPI library. The framework application need only implement a few functions in order to become an MPI parallel job. One feature of the framework is a set of classes that support the distribution of record keys, or key-value pairs, across the computation cluster. Key distribution allows for driving the test where all nodes have local access to the data. Key-value distribution has the advantage of running the job with the data source present only on a single node. Configuration of the job is managed with a simple text file that specifies the input data source, logging system (either files or a log server), and the number of data consumers assigned to each compute node. By using a configuration file, test scripts were simpler to invoke, the probability of error was reduced, and replication of the test scenario was implicitly provided. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 12 4.4 FPVTE – F INGERPRINT M ATCHING Validation Participants were provided imagery from publicly available datasets in the same format as the full evaluation datasets. These validation datasets were used to test implementation functionality before NIST performed the evaluation on a much larger dataset. Participants ran a validation test with the provided data on their own systems and provided NIST with the resulting candidate lists and scores from a set of pre-selected searches. NIST did the same on the machines used for the full evaluation to confirm that both NIST and the participant produced the same results. Validation also provided a way to make sure that participants were following the conventions of the evaluation, such as returning appropriate quality values and numbers of candidates, among others. Care was taken to make the validation process as simple as possible, to let the participant focus more on their submission than about intricacies specific to FpVTE. To assist, NIST provided a minimal version of the FpVTE test driver software used in the evaluation to aid participants through the expected calling structure of the evaluation API. Scripts were provided to compile the test driver and run the validation test. All a participant needed to do was create a properly named software submission, run the script, and submit the generated results file to NIST. Additionally, a build of the Biometric Evaluation Framework was provided, and many participants made use of features found in the framework in their API implementations. 4.5 Pre-Evaluation A timing test was performed to make sure no submission was in violation of the required time limits. First, a sample of the evaluation dataset was enrolled using ten processes per compute node and the timing of those enrollments were evaluated to make sure they did not exceed the average of three seconds per fingerprint. I/O time was not included during this step, as all data was passed to and from the FpVTE test driver in memory. If the submission passed the enrollment timing test, the full set of data was enrolled and finalized for matching. After enrollment, the next timing test was a random sample of mated and nonmated searches matched against the maximum enrollment set for each evaluation class. If the average search time on this sample test set, using a single process on each compute node, completed under the maximum time limit, the full search set of 30 000 searches was performed. The formula used to compute the total search time was: Ts = (t1 × b1 ) + t2 (1) where t1 = average identification stage one time b1 = number of blades required for identification stage one t2 = identification stage two time This formula allowed for a fair comparison between submissions that used one compute node for identification stage one versus others that may have used two or more compute nodes to store the enrollment set. The formula was vetted with the participants because it assumes that the enrollment set is evenly distributed across all identification stage one blades and that search time is fairly linear versus enrollment set size. Every attempt was made to use the same hardware and number of processes for every submission when reporting timing results. Additionally, the hardware used for timing was dedicated to the FpVTE test driver process with no other jobs running on the system except OS related processes. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 4.6 13 Receiving Submissions Every FpVTE submission received by NIST went through the FpVTE Validation and Pre-Evaluation processes before committing resources to complete the full evaluation. Figure 7 visualizes the submission process with a flowchart. Many submissions underwent the process multiple times due to defects in the initial submission. Potential FpVTE software submission defects that would require a resubmission include: . submission incorrectly signed or encrypted; . participant’s validation results differ from NIST’s; . software errors during evaluation; . maximum average time limit exceeded; . invalid API implementation. NIST provided no upper bound on the number of times that a participant could resubmit software in the event of a defect. Unfortunately, this created an unpredictably large amount of additional work for NIST, through managing submissions, helping participants debug, reporting status, and the like. Figure 6 details the number of non-debugging submissions received from participants over the course of the evaluation. The general trend of participants not achieving a “valid” submission on the first or second attempt ultimately played a part in forcing NIST to extend the original evaluation deadlines. Class A Class B Class C Number of Submissions 30 20 10 0 C D E F GH I J K L MO P Q S T U V C D E F GH I J K L MO P Q S T U V C D E F GH I J K L MOP Q S T U V Participant Round 1 Round 2 Round 3 Figure 6: Total number of non-debugging submissions received before the final deadline. At the conclusion of the evaluation, 733 FpVTE validation submissions were received by NIST, including validation submissions from participants who ultimately withdrew from the evaluation. For this total, a submission is considered a discrete transmission of non-debugging software to NIST that required action by a NIST employee, on a per-class basis. There were three rounds during which participants could make submissions. Participants were given an opportunity to update their submissions at the end of the first two submission rounds. Because FpVTE was a black-box evaluation, NIST C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 14 FPVTE – F INGERPRINT M ATCHING ultimately had no idea what changes to the submissions were made. A number of submissions majorly regressed in their lack of defects, so it can be anecdotally inferred that large portions or even entire submissions were changed by some participants. NIST imposed certain timing requirements on participants to ensure that the evaluation could be completed on NIST hardware in a reasonable amount of time, as well as to mimic a possible operational constraint. Several participants tried to tune their submissions to make use of the maximum amount of time, giving themselves a potential increase in overall accuracy. While the back-and-forth between NIST and the participants did increase the runtime of the evaluation, timing defects were the most understandable defect encountered in the evaluation and was primarily caused by NIST-based evaluation restrictions, not participant error. During the final submission, participants C, D, E, G, H, I, O, P, Q, and V provided one or more submissions that validated and ran to completion without NIST encountering defects other than timing restrictions. Participants C, E, G, I, O, and V met this requirement for all classes of participation in the final submission. Participant E was the only participant to achieve no defects other than timing restrictions, for every class during all three submission periods. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 15 Validation Pre-Evaluation NIST provides sample input data and evaluation source code. NIST runs the full evaluation over a set of 2 000 images from the evaluation dataset. Good Participants run their submissions linked to NIST source code over provided input data. NIST generates all templates for evaluation dataset. Crash Good NIST runs black-box submission over the same sample input data in the real evaluation environment. Perform timing test with subset of searches. Too Slow Full Evaluation Crash Good Bad Participants send blackbox submission and output from sample data to NIST. Run remaining (30 000) searches. Good Check results. Good Repeat searches for variable enrollment set sizes. Report results. Figure 7: Evaluation workflow. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 16 5 FPVTE – F INGERPRINT M ATCHING Two-Stage Matching The one-to-many identification step of FpVTE was divided into two distinct phases, under the expectation that storing the entire enrollment set in RAM on a single compute node would not have been possible. While this was true for many submissions, it did not hold true for all, especially with Class A data. Even if the submission’s enrollment set fit in the RAM of a single compute node, the two-stage match technique was still used. 5.1 Enrollment Set Partitioning NIST determined the amount of RAM needed to hold the entire enrollment set based on a size value returned per finger during enrollment. In some cases, unique forms of compression were employed, which allowed submissions to use significantly less RAM during identification than reported during enrollment. In other cases, submissions reported very small RAM requirements per finger, but underlying implementation details required significantly more RAM. This information was volunteered by participants and documented in Section 9. After the amount of required RAM was determined, NIST identified the number of compute nodes necessary to support the RAM requirements. Compute nodes had 192 GB of RAM each (see Subsection 4.2 for more information). Submissions were then invoked with a method asking them to “finalize” the set of enrollment templates for B compute nodes with a maximum of 192 GB of RAM per node. This gave submissions an opportunity to partition or subdivide the enrollment set into more manageable pieces on a per-compute node basis. 5.2 Identification — Stage One Once the enrollment set had been partitioned, or “finalized,” the next step was to perform searches on each of the partitions. First, the submission’s identification initialization method was called before the first stage of matching. It was expected that submissions would iterate over their enrollment set partition and load it into the RAM of the compute node for faster access. Some submissions spent much longer than anticipated in this initialization method, and may have performed some additional binning or pruning that wasn’t otherwise executed during the partitioning step. It’s important to note that the NIST evaluation compute nodes did not have swap enabled, and so the only memory that could have been allocated was physical RAM. After initialization, each search template was matched on each compute node specified during partitioning. To speed up this process, the FpVTE test driver forked into multiple processes. Under Linux, fork is implemented using copy-on-write pages, so as long as the submission’s child processes did not write to the RAM allocated during initialization, multiple identification processes could run in parallel with access to the enrollment set in RAM without fear of RAM being over allocated. Submissions were allocated a 4 GB RAM disk file system per compute node, where free-form data could be written. A RAM disk was used to avoid timing I/O as part of the identification process. The FpVTE test driver later persisted the data to a permanent storage device, and provided this data to the submission during the second stage of identification. If an submission failed to perform identification for any reason in the first stage of identification, it was marked as a miss in stage two, which increased false negative identification rate but slightly decreased false positive identification rate (Section 6). 5.3 Identification — Stage Two After all compute nodes had finished the first stage of identification, the second stage was invoked. The submission’s initialization method for the second stage was called, pointing the submission to the location of its partitioned enrollment sets (Subsection 5.1) and the RAM disk data from identification stage one. Then, each search template was submitted to stage two for final identification matching. This search method was to return a candidate list, not exceeding 100 candidates, with corresponding similarity scores in descending order (where the candidate at rank 1 was the most similar). Each call C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 17 to perform a search included a path to a directory rooted on a RAM disk file system, in which the FpVTE test driver placed the submission’s output from stage one (Subsection 5.2). There was no intended control on how the second stage of identification reached the final candidate list, but the search time limit imposed on submissions combined both stage one and stage two search times (see Section 4 for detailed information on timing constraints and calculations). Stage two could have been as simple as a sort of stage one results, or as complex as an additional level of matching involving templates. Based on the timing figures observed during the evaluation, both trivial and complex stage two implementations were used by FpVTE participants. Stage two took place on an individual compute node, though the pool of search templates may have been partitioned and searched on multiple compute nodes independently by the FpVTE test driver to increase throughput. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 18 FPVTE – F INGERPRINT M ATCHING 6 Metrics The detection error tradeoff (DET) characteristic curve [9] was the primary metric used for comparing accuracy in FpVTE. Specific points along the DET were examined for making comparisons between submissions. For this initial results report, the number of mate searches was limited to 10 000 and the number of nonmate searches was limited to 20 000 subjects, restraining the smallest error rate that could be used with statistical confidence to 10−3 . In addition to the accuracy of the submission, this report compares speed and computational resource usage. Comparisons between all submission are made and reported. 6.1 Accuracy Open-set identification algorithms can make two types of recognition error: . Search of a biometric sample of an individual not enrolled in the biometric enrollment set (a nonmated search) returns the biometric reference identifier(s) attributable to one or more enrolled person. This is considered Type I, or false alarm, because it returns a false identity. . Search of a biometric sample of an enrolled individual (a mated search) returns an incorrect enrolled identity. This is considered Type II, or miss, because it misses the correct identity. FpVTE quantified the accuracy of the open-set identification algorithms as follows: . False positive identification rate (FPIR), or Type I error rate, is the fraction of the nonmated searches where one or more enrolled identities are returned at or above threshold (T ) [4]. FPIR is a function of: the size of the enrollment set (N ), length of candidate lists (L), and score threshold (T ). In the general case, this can be summarized as FPIR(N, T, L) = Number of searches with any nonmates returned above threshold T on candidate list length L (2) Number of nonmated searches conducted and more precisely notated for this evaluation as PQ FPIR(T ) = q=1 H(dq1 − T ) Q (3) where Q is the number of searches performed for which there exists no mate in the enrollment set, dq1 is the highest similarity score reported by the algorithm for the q-th search. The function H(x) is the Heaviside step function H(x) = ( 0, if x < 0 1, if x ≥ 0 (4) . False negative identification rate (FNIR), or Type II error rate, is the fraction of the mated searches where the enrolled mate is outside the top R rank or comparison score is below threshold (T ). FNIR is a function of: the size of the enrollment set (N ), length of candidate lists (L), score threshold (T ), and the number of top candidates being considered (R). This is summarized in the general case as FNIR(N, R, T, L) = Number of mates outside top R ranks or below threshold T on candidate list length L Number of mated searches conducted C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology (5) FPVTE – F INGERPRINT M ATCHING 19 and is defined formally for this evaluation as FNIR(T ) = 1 − P L 1 XX Ipr [1 − H(dpr − T )] P p=1 r=1 (6) where P is the number of searches performed for which there exists a mate in the enrollment set, dpr is the r-th lowest similarity score reported by the algorithm for the p-th search, and Ipr is 1 if the identity of the r-th candidate is the same as the identity of search p, or 0 otherwise. Note that FNIR computation does not care about the cause of a miss: failure to correctly identify a sample (e.g., due to poor quality), failure to extract a template, failure to generate a comparison score, and software crashes are all dealt with similarly. The terms “hit rate,” “reliability,” and “sensitivity” that have been mentioned in some literature on automated fingerprint identification systems (AFIS) [8, 16] are just the complement of FNIR, computed as 1 − FNIR. Another widely used accuracy metric is cumulative match characteristic (CMC), which is the fraction of the mated searches where the enrolled mate is at rank R or better, regardless of its comparison score. CMC is a special case of FNIR, or more precisely, hit rate, when the constraint on threshold is removed, as shown in Equation 7. CMC(N, L, R) = 1 − FNIR(N, L, T = 0, R) (7) Rank-one hit rate, CMC(N, L, R = 1), is the most common accuracy metric reported in academic and AFIS-related literature. While CMC is reported for the tested submissions, it is an inadequate accuracy metric because its makes strong or weak hits indistinguishable by ignoring similarity scores, and does not report Type I errors. 6.2 DET Plots DET characteristic curves are the primary accuracy metric for offline testing of biometric recognition algorithms. Each point on a DET curve exhibits the false positive identification and false negative identification rates associated with a certain threshold value. The DET curve spans the entire range of possible threshold values, which is normally the range of the comparison scores. To reveal the dependence of FNIR and FPIR at a fixed threshold, the DET curves are connected at points where FNIRs and FPIRs are observed at the same threshold values. As it is conventional, DET curves are presented for FpVTE submission. In a DET curve, Type I error rates are plotted on the x-axis and Type II error rates are plotted on the y-axis, giving uniform treatment to both types of error. Both axes use a logarithmic scale, which spreads out the plot and better distinguishes different well-performing systems. When calculating FPIR and FNIR, all ranks were considered (L = R = 100). 6.3 Failure to Extract or Match a Template Failure to extract is the fraction of images for which a template is not generated. Template generation can fail for the enrollment sample or the search sample. In both cases, failure to extract a template is included as a miss in the computation of FNIR (see section 6.1). Additionally, recognition algorithms fail to execute one-to-many searches to produce comparison scores. The result is that a valid candidate list is not produced. Such failures might be voluntary (e.g., refusal to process a poor quality image) or involuntary (e.g., software crashes). Either way, it is an undesirable behavior, and should be included in computation of recognition errors, particularly to allow for fair comparison of submissions. FpVTE treated such failure cases as a miss and added them to the Type II errors. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 20 FPVTE – F INGERPRINT M ATCHING Right Hand Left Hand 1 2 3 4 5 11 13 6 7 8 9 10 12 14 Table 6: Table showing which finger positions were swapped when the images were flipped for nonmate searches. 6.4 Figure 9: Example of flipped single-index finger image. Computational Efficiency Another aspect of performance is the computational resources required by a submission. This report includes a comparison of template generation times, one-to-many search times, and template sizes for the submissions, along with their accuracy at a set FPIR point of 10−3 . The timing numbers are based on data samples for which all submissions were run under identical conditions on the same hardware. Ground Truth Errors The nonmate search errors were resolved by flipping fingerprint images on the vertical axis for nonmated searches. In order to have the correct topography of the finger, left/right hand labelings and finger positions were reversed after flipping. Table 6 shows how the finger positions were swapped to keep the correct finger topography. Example images are shown in Figure 9, Figure 10, and Figure 11. 0 There were two types of ground truth errors that had to be resolved for FpVTE datasets. The first involved nonmate searches that had an unknown mate in the enrollment set. These errors would wrongly increase FPIR (see Section 6). The second involved mated searches that did not match the presumed mate or had other unknown mates in the enrollment set. Either of these mated search errors could wrongly increase FNIR. An example of the effect on a DET curve is shown in Figure 8. Once the errors show up at a certain threshold, the error rate sharply increases and remains erroneously high. FNIR 1 6.5 0 FPIR 1 Figure 8: Example of a DET showing a spike in False Positive Identification Rate. This is usually a sign of ground truth errors in nonmate searches. The mated search errors had to be resolved through manual inspection. A large cause of the unknown mates resulted from FpVTE using data from multiple sources. These ground truth errors had to be detected by examining the unexpected high-scoring alleged nonmates produced by the submissions. The results from the submissions were grouped together to determine which unexpected high-scoring alleged nonmates needed manual inspection. The first step was to look at searches where all the submissions had an unexpected high-scoring alleged nonmate above a certain threshold. Next, cases were examined for which only some of the submissions had a high-scoring alleged nonmate. After examining these cases, if the majority were true mates, the thresholds used were decreased and the process repeated until very few or no more true mates were found. Not all submissions were used for this process as some produced results that would have required too much manual work to inspect all of the potential errors produced. This process was repeated for low scoring alleged mates. The mate searches with low scores were examined to determine if the alleged mate was truly a mate or if there was a ground truth error. Very low quality images were not removed from the datasets. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Figure 10: Example of flipped left slap image. Note that the flipped image was used as a right slap. Figure 11: Example of flipped right slap image. Note that the flipped image was used as a left slap. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 21 22 7 FPVTE – F INGERPRINT M ATCHING Accuracy Results Subsections 7.1 through 3 show the identification accuracy results, from round 3 submissions, for all three classes of participation. The sections include plots sorted by rank for FNIR at a fixed FPIR of 10−3 , Detection Error Tradeoff (DET) curves showing accuracy over a range of threshold values, Cumulative Match Characteristic (CMC) curves showing accuracy over a range of candidate list ranks, and tables with FNIR values at a fixed FPIR. A complete set of full-size DET curves for each participant are included in Appendix A. Appendix B shows DET and CMC plots for all submissions and classes grouped on a single page. This section (7) is followed by sections showing accuracy tradeoff results. Section 8 shows FNIR compared to search time statistics. Appendix C and Appendix D have a complete set of tables for search time statistics. Appendix E shows the progression of timing and results from the second to third round of submissions. Section 9 shows FNIR compared to computational resources such as RAM usage and enrollment (i.e., template creation) times. Appendices F to H have detailed tables on templates sizes and creation times. These results are coalesced in Section 10, with tables combining ranked results for each category (FNIR, search time, RAM, enrollment time). Appendix I has the complete set of tables for ranked results in all classes of participation. Appendix K plots relative comparisons of FNIR, RAM usage, template creation times, and search times. Section 11 combines results across classes of participation showing how accuracy varied based on the number of fingers available for searching. Appendix J plots relative comparisons, by class, for each search set used in FpVTE. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 7.1 Class A 7.1.1 Single-Index Finger Identification 23 The accuracy of single-finger identification is shown with rank-sorted FNIR points (based on right index finger results) in Figure 12, DET curves in Figure 13, CMC plots in Figure 14, and tables of FNIR points at a fixed FPIR in Tables 7 through 8. Some observations for Class A single-finger identification include (all FNIR values are at FPIR =10−3 ): . The most accurate submissions were D, Q, I, V and L2. . The right index finger was more accurate than the left index finger. . The most accurate submission D achieved a FNIR of 1.97% for the left index finger and 1.9% for the right index finger searched against an enrollment set of 100 000 subjects. . There is a measurable accuracy gap between the top performers and the next level of performers. . The CMC plots in Figure 14 are not as flat as the other classes. This indicates that while most mates appear within the top three candidates on the list, there are some that appear further down the list when using single-finger identification. ● ● 0.20 ● ● FNIR @ FPIR =10−3 ● ● ● 0.10 ● ● ● ● ● 0.05 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.02 ● ● ● ● ● ● ● ● ● ● ● ● ● ● D1 D2 I2 Q2 I1 Q1 V2 V1 L2 S1 S2 L1 T2 E2 E1 J2 O2 K1 K2 J1 O1 F2 G2 G1 F1 U1 U2 P2 C2 C1 P1 H1 H2 T1 M2 M1 Submission ● Left Index ● Right Index Figure 12: Rank-sorted FNIR @ FPIR = 10−3 for Class A — Single Index Finger searching 30 000 subjects against 100 000 subjects. Submissions “1” and “2” from round 3. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 24 0.5000 0.2000 ● ● ● ●● ● ● ● ● ● I C ● ●● ● ● ● P ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● J D Left Index (1) ● ●● ● ● ● ● ● Q ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● E K S ● ● Left Index (2) ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● T L ● ● F Right Index (1) ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● G ● M U ● Right Index (2) ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● 0.5000 ● ● ● ● 0.1000 0.0500 0.0200 0.0100 ● 0.0050 ● ● 0.5000 ● ● ● ● 0.0050 0.0100 0.0200 ● ● ● ● 0.0500 0.1000 0.2000 ● ● ● ● ● ● ●●●● 0.0005 0.0010 0.0020 ● ● ● ● ● ● ●●●● 0.0500 0.1000 0.2000 0.2000 ● ● ●● 0.0050 0.0100 0.0200 ● ● ●● ●● ● ● 0.0050 0.0100 0.0200 0.5000 False Positive Identification Rate 0.0005 0.0010 0.0020 0.1000 ● 0.0500 0.1000 0.2000 0.0500 0.0200 0.0100 0.0050 0.0005 0.0010 0.0020 0.5000 ● ●● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● H O ● ● V ●● ●● ● ● ● ● ● ● ● ● Figure 13: DET for Class A — Single Index Finger searching 30 000 subjects against 100 000 subjects. Submissions “1” and “2” from round 3. False Negative Identification Rate C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Miss Rate 5 10 P I C 15 Q J D 5 10 S K E Left Index (2) 15 Rank T L F Right Index (1) 5 10 U M G Right Index (2) 15 V O H Figure 14: CMC for Class A — Single Index Finger searching 30 000 subjects against 100 000 subjects. Submissions “1” and “2” from round 3. 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 Left Index (1) FPVTE – F INGERPRINT M ATCHING 25 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 26 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Sub. # FNIR @ FPIR = 10−3 1 30 2 31 0.1337 1 1 0.0197 2 1 0.0197 1 16 0.0745 2 15 0.0723 1 25 0.1111 2 22 0.1082 1 24 0.1089 2 23 0.1086 1 32 0.1576 2 33 0.1607 1 7 0.0257 2 8 0.0278 1 18 0.0786 2 14 0.0712 1 21 0.0883 2 20 0.0875 1 11 0.0625 2 9 0.0351 1 35 0.2995 2 34 0.2921 1 19 0.0818 2 17 0.0766 1 29 0.1308 2 28 0.1272 1 3 0.0222 2 4 0.0226 1 10 0.0571 2 12 0.0650 1 36 NA 2 13 0.0685 1 27 0.1218 2 26 0.1178 1 6 0.0253 2 5 0.0252 0.1335 Table 7: Tabulation of results for Class A — Left Index, with an enrollment set size of 100 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. Participant Letter C D E F G H I J K L M O P Q S T U V Sub. # FNIR @ FPIR = 10−3 1 30 2 29 0.1124 1 1 0.0190 2 1 0.0190 1 15 0.0630 2 14 0.0624 1 25 0.0933 2 22 0.0903 1 24 0.0910 2 23 0.0909 1 32 0.1230 2 33 0.1249 1 5 0.0215 2 3 0.0214 1 20 0.0708 2 16 0.0643 1 18 0.0682 2 19 0.0685 1 12 0.0505 2 9 0.0295 1 36 0.2615 2 35 0.2526 1 21 0.0776 2 17 0.0675 1 31 0.1133 2 28 0.1100 1 6 0.0218 2 3 0.0214 1 10 0.0442 2 11 0.0503 1 34 0.1929 2 13 0.0562 1 26 0.0996 2 27 0.1007 1 8 0.0223 2 7 0.0222 0.1132 Table 8: Tabulation of results for Class A — Right Index, with an enrollment set size of 100 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . The number to the left of a value provides the value’s columnwise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 7.1.2 27 Two-Index Finger Identification The accuracy of two-finger identification is shown with rank-sorted FNIR points in Figure 15, DET curves in Figure 16, CMC plots in Figure 17, and a table of FNIR points at a fixed FPIR in Table 9. Some observations for Class A two-finger identification include (all FNIR values are at FPIR =10−3 ): . The most accurate submissions were Q, V, D, and I. . The most accurate submission Q achieved a FNIR of 0.27% searched against an enrollment set of 1.6 million subjects. . The accuracy gap between the top performers and the second tier, while still measurable, was not as large as singlefinger identification. . Two-finger identification was far superior to single-finger identification and scaled to much larger enrollment sets. . The CMC plots in Figure 17 flatten out faster than in single-finger identification. In most cases, the mate is within the top three candidates of the list or doesn’t appear at all. There were a few submissions that extend down to the top five to ten candidates. ● ● ● 0.050 ● FNIR @ FPIR =10−3 ● 0.020 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.010 ● ● 0.005 ● ● ● ● ● ● ● Q1 Q2 V2 D1 D2 I2 V1 I1 L2 J1 J2 L1 S2 E2 E1 O2 O1 S1 K2 G2 P2 U1 U2 K1 T2 C1 P1 C2 F1 F2 G1 H2 H1 M2 M1 T1 Submission ● Left and Right Index Figure 15: Rank-sorted FNIR @ FPIR = 10−3 for Class A — Two Index Fingers searching 30 000 subjects against 1 600 000 subjects. Submissions “1” and “2” from round 3. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 28 0.5000 0.2000 0.1000 0.0500 ● ● 0.0200 0.0100 0.0050 ● ● ● I C ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● D J ● ●● ● ● ● ● ● ●● S ● K E Left and Right Index (1) ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● 0.5000 0.0020 0.0010 0.0005 ● ● Q ● ● ● ● ● ● ● F ● L T ● ● ● ● ● ● ● ● Left and Right Index (2) ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● G M U ● ● ● ● ● ●● ● ● ● 0.0500 0.1000 0.2000 ● ● 0.0050 0.0100 0.0200 ● ●● 0.0005 0.0010 0.0020 P ● ●● 0.0500 0.1000 0.2000 ● ● ●● 0.0050 0.0100 0.0200 False Positive Identification Rate 0.0005 0.0010 0.0020 0.5000 ● ● 0.0500 0.1000 0.2000 0.2000 0.1000 0.0500 ● 0.0050 0.0100 0.0200 ● ● 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0005 0.0010 0.0020 0.5000 ● ● 0.5000 ● ● ● ● ● ● ●● ● ● ● H O V ● ● ● ● ● ● ● Figure 16: DET for Class A — Two Index Fingers searching 30 000 subjects against 1 600 000 subjects. Submissions “1” and “2” from round 3. False Negative Identification Rate C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Miss Rate 5 10 P I C 15 Q J D 5 10 S K E 15 Rank T L F Left and Right Index (2) 5 10 U M G 15 V O H Figure 17: CMC for Class A — Two Index Fingers searching 30 000 subjects against 1 600 000 subjects. Submissions “1” and “2” from round 3. 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 Left and Right Index (1) FPVTE – F INGERPRINT M ATCHING 29 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 30 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Sub. # FNIR @ FPIR = 10−3 1 26 0.0368 2 28 0.0374 1 4 0.0030 2 4 0.0030 1 15 0.0207 2 14 0.0202 1 29 0.0386 2 30 0.0412 1 31 0.0515 2 20 0.0311 1 33 0.0686 2 32 0.0684 1 8 0.0058 2 4 0.0030 1 10 0.0143 2 10 0.0143 1 24 0.0360 2 19 0.0286 1 12 0.0146 2 9 0.0072 1 35 NA 2 34 NA 1 17 0.0229 2 16 0.0214 1 27 0.0370 2 21 0.0333 1 1 0.0027 2 1 0.0027 1 18 0.0281 2 13 0.0195 1 36 NA 2 25 0.0366 1 22 0.0336 2 23 0.0358 1 7 0.0034 2 3 0.0028 Table 9: Tabulation of results for Class A — Left and Right Index, with an enrollment set size of 1 600 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 7.2 31 Class B The accuracy results for Class B, which included four-finger, eight-finger, and ten-finger identification, are shown with rank-sorted FNIR points (based on ten finger results) in Figure 18, DET curves in Figure 19, CMC plots in Figure 20, and tables of FNIR points at fixed FPIR in Tables 10 through 13. Some observations for Class B include (all FNIR values are at FPIR =10−3 ): . The most accurate submissions with all ten fingers were I, Q, and D, with FNIRs ranging from 0.09% to 0.2%. . The next level of submissions were V, E, L, J, O, and G, with FNIRs ranging from 0.24% to 0.4%. The separation between the top performers and next level of performers was noticeably lower as more fingers are used. . Right slaps (FNIR = 0.45%, I2) were more accurate than left slaps (FNIR = 0.94%, I2). . Four-finger identification (FNIR = 0.45%) performed worse than two-finger identification (FNIR = 0.27%), as reported in Subsubsection 7.1.2. Two potential causes for this result are that the submissions in Class B had to perform segmentation as part of the feature extraction process, and possibly a variation in image quality between the datasets. Further study will be needed to determine the primary reason four-finger slaps performed worse than two index fingers. . The matching accuracy improved significantly going from four fingers to eight fingers. . The best submission I2 achieved FNIRs as follows: left slap (0.94%), right slap (0.45%), left and right slaps (0.15%), and identification slaps (0.09%). . The CMC plots in Figure 20 generally show very flat responses. This indicates that for most submissions, the mate is within the top three candidates on the list or the mate doesn’t appear on the list at all. ● FNIR @ FPIR =10−3 0.050 ● ● ● ● ● ● 0.020 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.005 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.010 ● ● 0.100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● D2 I1 0.002 ● ● ● ● 0.001 ● I2 Q1 Q2 D1 E2 V2 V1 L1 J2 L2 O2 G2 O1 E1 J1 G1 U1 S1 S2 U2 H1 H2 M2 M1 F1 F2 C2 C1 Submission ● Identification Flats ● Left and Right Slap ● Left Slap ● Right Slap Figure 18: Rank-sorted FNIR @ FPIR = 10−3 for Class B — Left Slap, Right Slap, Left and Right Slap, and IDFlats searching 30 000 subjects against 3 000 000 subjects. Submissions “1” and “2” from round 3. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 32 0.2000 ● ● ● ● ● ● ● ● 0.0020 ● ● ● 0.1000 ● ● 0.0500 0.0200 0.0100 ● ● ● ● ● M ● G ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● Left Slap (1) Left Slap (2) ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ● ● ●● ● ●● ●● ●● ● ● ● ● ● Right Slap (1) Right Slap (2) C ● ● ● ● ● ● ● ● ● ● H ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● D I S ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Left and Right Slap (1) Left and Right Slap (2) ● ● 0.0500 0.0050 0.0020 0.0010 ● ● ● ● ● ● ● ● ● ● ● ● 0.0020 ● ● 0.5000 0.0005 ● ● ● ● Q ● ● ● ● ●● ● ● ● ● ● ● ● E J ● ● U ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ●● ● ● Identification Flats (1) Identification Flats (2) ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● 0.0020 ● ● ● ● ●● 0.0200 ● ● ● 0.0500 ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● F L ● ● ● ● V ● ● 0.0200 O ● ● 0.5000 ● ●● 0.2000 ● 0.0200 0.2000 ● ● ●● 0.0050 0.1000 ● ● ● 0.1000 ●● 0.0010 ● 0.0010 0.0500 ● ● 0.0100 0.0200 ●● ● ● ● ● 0.0010 ● ● ● 0.0005 ● ● ● ● ● ● 0.0500 ● 0.2000 ● ● ● 0.0100 ● 0.0050 0.0100 ● ●● 0.0100 0.0050 ● ● 0.1000 ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● 0.2000 ● ● 0.1000 0.0020 0.0005 False Positive Identification Rate 0.0050 0.0010 0.0005 0.0005 0.5000 Figure 19: DET for Class B — Left Slap, Right Slap, Left and Right Slap, and IDFlats searching 30 000 subjects against 3 000 000 subjects. Submissions “1” and “2” from round 3. False Negative Identification Rate C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Miss Rate 5 10 O G M 15 Q H C Right Slap (1) Right Slap (2) 5 Rank 10 S I D 15 Left and Right Slap (1) Left and Right Slap (2) U J E Identification Flats (1) Identification Flats (2) 5 10 V L F 15 Figure 20: CMC for Class B — Left Slap, Right Slap, Left and Right Slap, and IDFlats searching 30 000 subjects against 3 000 000 subjects. Submissions “1” and “2” from round 3. 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 Left Slap (1) Left Slap (2) FPVTE – F INGERPRINT M ATCHING 33 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 34 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 22 0.0654 2 21 0.0647 1 6 0.0163 2 5 0.0142 1 13 0.0259 2 7 0.0187 1 29 0.1684 2 28 0.1681 1 18 0.0371 2 17 0.0325 1 23 0.0998 2 24 0.1008 1 4 0.0116 2 1 0.0094 1 15 0.0287 2 10 0.0236 1 16 0.0288 2 14 0.0276 1 30 0.1736 2 27 0.1634 1 12 0.0257 2 11 0.0254 1 2 0.0098 2 3 0.0099 1 25 0.1089 2 26 0.1133 1 20 0.0500 2 19 0.0461 1 9 0.0192 2 8 0.0190 Table 10: Tabulation of results for Class B — Left Slap, with an enrollment set size of 3 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 24 0.0403 2 23 0.0392 1 6 0.0072 2 2 0.0052 1 13 0.0151 2 7 0.0083 1 29 0.1222 2 28 0.1220 1 18 0.0212 2 16 0.0198 1 25 0.0641 2 26 0.0647 1 5 0.0058 2 1 0.0045 1 14 0.0156 2 10 0.0126 1 15 0.0167 2 17 0.0202 1 30 0.1259 2 27 0.1155 1 12 0.0142 2 11 0.0132 1 3 0.0057 2 3 0.0057 1 21 0.0369 2 22 0.0381 1 19 0.0266 2 20 0.0273 1 8 0.0106 2 9 0.0110 Table 11: Tabulation of results for Class B — Right Slap, with an enrollment set size of 3 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . The number to the left of a value provides the value’s columnwise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 30 NA 2 29 NA 1 6 0.0031 2 5 0.0024 1 15 0.0063 2 10 0.0049 1 28 0.0910 2 26 0.0901 1 18 0.0106 2 17 0.0084 1 23 0.0349 2 24 0.0361 1 3 0.0022 2 1 0.0015 1 16 0.0068 2 9 0.0047 1 12 0.0054 2 14 0.0062 1 27 0.0904 2 25 0.0882 1 13 0.0057 2 11 0.0051 1 2 0.0021 2 3 0.0022 1 21 0.0160 2 22 0.0190 1 20 0.0139 2 19 0.0124 1 7 0.0036 2 7 0.0036 Table 12: Tabulation of results for Class B — Left and Right Slap, with an enrollment set size of 3 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. 35 Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 30 NA 2 29 NA 1 6 0.0020 2 2 0.0012 1 16 0.0043 2 7 0.0024 1 27 0.0591 2 27 0.0591 1 18 0.0062 2 14 0.0040 1 23 0.0203 2 24 0.0204 1 2 0.0012 2 1 0.0009 1 17 0.0049 2 11 0.0033 1 10 0.0031 2 11 0.0033 1 26 0.0543 2 25 0.0515 1 15 0.0041 2 13 0.0035 1 2 0.0012 2 2 0.0012 1 20 0.0108 2 21 0.0136 1 19 0.0099 2 22 0.0141 1 9 0.0027 2 7 0.0024 Table 13: Tabulation of results for Class B — Identification Flats, with an enrollment set size of 3 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 36 FPVTE – F INGERPRINT M ATCHING 7.3 Class C The accuracy results for Class C, which included ten-finger identification for plain and rolled impression types using scanned ink and livescan data, are shown with rank-sorted FNIR points (based on ten-finger rolled-to-rolled impression results) in Figure 21, DET curves in Figure 22, CMC plots in Figure 23, and tables of FNIR points at a fixed FPIR in Tables 14 through 16. Some observations for Class C include (all FNIR values are at FPIR =10−3 ): . The most accurate submissions were I, Q, D, and V (ten-finger rolled-to-rolled), with FNIRs ranging from 0.1% to 0.19% for ten-finger plain impressions and ten-finger rolled impressions. . There was not a clear difference between ten-finger plain-to-plain and ten-finger rolled-to-rolled results. This is a bit of a surprise, as the ten-finger plain data had to be segmented by the submission and the ten-finger rolled data did not. It also is a surprise after the observations of lower accuracy seen in Class B four-finger slaps versus the Class A two index finger accuracy (Subsection 7.2). It might be interesting to perform a similar Class C test with only four-finger slaps to see if the results are similar to Class B four-finger slap results. . The second level of performers were E2, J, O, and V (ten-finger plain) with FNIRs ranging from 0.25% to 0.50% for ten-finger plain-to-plain and ten-finger rolled-to-rolled impressions. . The best performers were able to handle both rolled and plain impression images with little variation in FNIR, there was a slight decrease when comparing plain-to-rolled impressions. . Similar to Class B, the Class C CMC plots in Figure 23 are very flat, indicating the mate is within the top two positions on the candidate list or it is completely missed. ● ● ● ● ● ● 0.200 ● 0.100 FNIR @ FPIR =10−3 ● ● ● 0.050 0.020 ● 0.010 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.005 ● ● ● ● ● ● ● ● ● ● ● ● ● 0.002 ● ● ● ● ● I1 I2 ● ● ● ● ● ● ● ● ● ● ● ● ● Q2 D1 Q1 V1 D2 V2 J2 O2 O1 E2 J1 L2 C2 C1 L1 E1 H2 H1 G2 U2 U1 G1 F1 F2 M2 M1 S1 S2 Submission ● Ten−Finger Plain−to−Plain ● Ten−Finger Plain−to−Rolled ● Ten−Finger Rolled−to−Rolled Figure 21: Rank-sorted FNIR @ FPIR = 10−3 for Class C — Ten-Finger plain-to-plain, rolled-to-rolled, and plain-to-rolled searching 30 000 subjects against 5 000 000 subjects. Submissions “1” and “2” from round 3. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology ● ● 0.0020 0.0010 0.0005 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● O ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● Q H ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● I ● ● ● ● S ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● False Positive Identification Rate ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● U ● J ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● V L ● F ● ● ● ● ● ● ● ● 0.5000 0.0050 False Negative Identification Rate Figure 22: DET for Class C — Ten-Finger plain-to-plain, rolled-to-rolled, and plain-to-rolled searching 30 000 subjects against 5 000 000 subjects. Submissions “1” and “2” from round 3. ● 0.0200 0.0100 0.0050 ● ● 0.2000 0.1000 0.0500 0.5000 0.0005 ● 0.0010 ● ● ● 0.0020 ● 0.0050 ● 0.0200 ● 0.0100 G 0.0500 0.0020 0.0010 0.0005 0.1000 0.0200 0.0100 0.0050 0.2000 ● 0.0005 ● 0.0010 ● 0.0020 ● 0.0050 ● 0.0100 ● 0.0200 E 0.0010 ● 0.0500 D 0.1000 C 0.2000 ● 0.5000 0.2000 ● 0.1000 ● 0.0500 0.5000 ● 0.0005 ● 0.0020 ● 0.0100 M 0.0200 0.5000 0.0500 Ten−Finger Plain−to−Rolled (1) Ten−Finger Plain−to−Rolled (2) 0.1000 Ten−Finger Rolled−to−Rolled (1) Ten−Finger Rolled−to−Rolled (2) 0.2000 Ten−Finger Plain−to−Plain (1) Ten−Finger Plain−to−Plain (2) FPVTE – F INGERPRINT M ATCHING 37 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 38 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 5 E L F Ten−Finger Plain−to−Rolled (1) Ten−Finger Plain−to−Rolled (2) D J Ten−Finger Rolled−to−Rolled (1) Ten−Finger Rolled−to−Rolled (2) C I Ten−Finger Plain−to−Plain (1) Ten−Finger Plain−to−Plain (2) M H 10 G 5 V 15 U 10 S 5 Q 15 O 10 Rank 15 Figure 23: CMC for Class C — Ten-Finger plain-to-plain, rolled-to-rolled, and plain-to-rolled searching 30 000 subjects against 5 000 000 subjects. Submissions “1” and “2” from round 3. Miss Rate C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 30 NA 2 24 0.0711 1 6 0.0015 2 2 0.0011 1 14 0.0088 2 13 0.0048 1 25 0.0734 2 25 0.0734 1 23 0.0368 2 20 0.0276 1 20 0.0276 2 19 0.0275 1 4 0.0013 2 1 0.0010 1 12 0.0047 2 10 0.0027 1 16 0.0102 2 15 0.0095 1 28 0.0934 2 27 0.0826 1 9 0.0025 2 10 0.0027 1 2 0.0011 2 4 0.0013 1 22 0.0311 2 29 0.1680 1 18 0.0163 2 17 0.0155 1 7 0.0024 2 7 0.0024 Table 14: Tabulation of results for Class C — Ten-Finger Plain-toPlain, with an enrollment set size of 5 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. 39 Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 16 0.0094 2 15 0.0085 1 4 0.0015 2 7 0.0018 1 18 0.0106 2 12 0.0050 1 25 0.0536 2 25 0.0536 1 24 0.0447 2 21 0.0333 1 20 0.0201 2 19 0.0199 1 1 0.0013 2 2 0.0014 1 13 0.0051 2 9 0.0033 1 17 0.0097 2 14 0.0083 1 28 0.0783 2 27 0.0716 1 11 0.0034 2 9 0.0033 1 5 0.0017 2 2 0.0014 1 29 0.0860 2 30 0.2462 1 23 0.0358 2 22 0.0351 1 5 0.0017 2 8 0.0019 Table 15: Tabulation of results for Class C — Ten-Finger Rolled-toRolled, with an enrollment set size of 5 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 40 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Sub. # FNIR @ FPIR = 10−3 1 17 0.0149 2 18 0.0169 1 6 0.0028 2 3 0.0018 1 16 0.0137 2 12 0.0056 1 27 0.2514 2 27 0.2514 1 24 0.0649 2 23 0.0521 1 20 0.0291 2 19 0.0285 1 2 0.0014 2 1 0.0011 1 13 0.0071 2 7 0.0034 1 15 0.0136 2 14 0.0129 1 30 0.3067 2 27 0.2514 1 10 0.0041 2 8 0.0036 1 4 0.0020 2 5 0.0022 1 25 0.1017 2 26 0.2366 1 22 0.0378 2 21 0.0295 1 11 0.0052 2 9 0.0039 Table 16: Tabulation of results for Class C — Ten-Finger Plain-to-Rolled, with an enrollment set size of 5 000 000. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. FNIR was computed at the score threshold that gave FPIR = 10−3 . The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 8 41 Accuracy/Search Time Tradeoff This section examines the tradeoff between FNIR and the amount of time each submission needed to perform identification searches. The only time restriction placed on the submissions were they must complete searches before a maximum time limit (Class A: 500 seconds, Class B/C: 90 seconds). NIST allowed for and encouraged two submissions per round. The intent of this decision was to demonstrate the tradeoff between accuracy and speed, but there was no requirement that the same basic algorithm be used for the intended “fast” and “slow” submission. It was possible that two completely different algorithmic approaches were used by a participant, which could introduce other factors when comparing accuracy to search time for a given participant. As such, the two submissions are simply labeled “1” and “2” in this report. The timing tables shown in this section are based on the detailed timing tables in Appendix C that show the timing for each stage of identification. In addition to the accuracy/timing tradeoffs plots and timing tables in this section, Appendix E contains a full set of tables that show timing changes that occurred between the last two rounds of submissions. Those tables provide more data to analyze tradeoffs between search time and accuracy. Again, the algorithm could have changed from one round of submissions to the next, but it is generally assumed that the basic algorithmic approach stayed the same, while “controls” were tweaked to improve accuracy. One general observation noticed across classes of participation was that increased search times are not a guarantee of increased accuracy. This turned out to be inconsistent across the test. In general, most submissions obtain some improvement in accuracy with increased search times, but most gains are modest, and a few cases had no gain or a slight loss in accuracy. There are other cases where submissions decreased search time yet still improved accuracy. It might be useful in future evaluations to encourage competition over extremely fast searches and see which submissions have a search time vs accuracy advantage, as opposed to tweaking for maximum accuracy, as was done in the current FpVTE protocol. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 42 8.1 FPVTE – F INGERPRINT M ATCHING Class A Tabulated comparisons of identification times for index finger identification submissions are shown in Tables 17 and 18. The search times shown in these tables are from the “Total/One” column in Tables 24 through 29 included in Appendix C. For reference, the FNIR values from Section 7 are reprinted to the right of the search times. Class A tables were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. The tables were used to create scatter plots showing accuracy, search times, and search template creation times. Those plots are shown in Figures 24 through 29. Some observations for Class A identification times include: . For single index fingers, there was essentially no accuracy improvement observed with larger search times. . For two index fingers, there was improvement shown by some participants with increased search time. The overall benefit might depend on the application. . It is difficult to compare single-finger identification results to two-finger identification results here, as the enrollment set sizes used were not the same (100 000 and 1.6 million, respectively). . The number of processes running (one or ten) didn’t appear to have a major effect on throughput for Class A single index fingers, and only a slight increase in processing time was observed for two-finger searches. See Appendix C and Appendix D for complete details. . Tables 50 through 52 in Appendix E and Figures 30 through 35 show differences between the last two rounds of submissions. Most submissions lowered FNIR for single finger but required longer search times. The results for two index fingers were not as consistent. Most lowered FNIR, but some had a significant increase in search time for little gain in FNIR. While it is not known what changes were made between submissions, there is some indication that high accuracy can be achieved with some of the fast submissions. The absolute best accuracy is achieved by slightly slower submissions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 43 Left Index (Less Than 20−Second Searches) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.50 ● T1 ● ● FNIR @ FPIR =10−3 M1 M2 0.20 C1 ● 0.10 0.05 ● C2 H1 ● H2 ● ● P2 P1 ● F1 ● U1 ● G1 ● ● F2 K2 ●● K1 E1 ●● O1 ● ● J1 ● O2 ● J2 L1 E2 ● ● T2 ● S1 ● L2 I1 ● ● V1 0.02 ● ● D1 0.01 5 ● Q2 Q1 10 15 Median Search Time − One Process (seconds) Figure 24: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and median search time (less than 20 seconds) for a single process for Class A — Left Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 17 and search template creation times can be found in Table 72 in Appendix H. Left Index (Greater Than or Equal To 20−Second Searches) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 FNIR @ FPIR =10−3 0.50 0.20 ● ● U2 0.10 G2 ● S2 0.05 ● I2 0.02 0.01 ● V2 ● D2 10 20 30 40 50 Median Search Time − One Process (seconds) 60 Figure 25: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and median search time (greater than or equal to 20 seconds) for a single process for Class A — Left Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 17 and search template creation times can be found in Table 72 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 44 FPVTE – F INGERPRINT M ATCHING Right Index (Less Than 20−Second Searches) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.50 FNIR @ FPIR =10−3 ● 0.20 ● M1 M2 ● T1 H1 ● ● P1 C1 H2 ● ● ● ● C2 P2 ● F1 ● F2 O1 ● ● J1 ● O2 ● ● J2 E1 ● 0.05 L1 0.10 G1 ● U1 ● K2 ●● K1 ● E2 ● T2 ● S1 ● L2 ● 0.02 V1 0.01 5 Q1 ● I1 ●● Q2 ● D1 10 15 Median Search Time − One Process (seconds) Figure 26: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and median search time (less than 20 seconds) for a single process. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 17 and search template creation times can be found in Table 73 in Appendix H. Right Index (Greater Than or Equal To 20−Second Searches) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 FNIR @ FPIR =10−3 0.50 0.20 0.10 ● U2 ● G2 0.05 ● S2 ● 0.02 0.01 ● D2 10 I2 20 30 40 50 Median Search Time − One Process (seconds) ● V2 60 Figure 27: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and median search time (greater than or equal to 20 seconds) for a single process for Class A — Right Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 17 and search template creation times can be found in Table 73 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 45 Left and Right Index (Less Than 20−Second Searches) 0 1 2 3 4 5 0.100 0.050 ● G1 FNIR @ FPIR =10−3 K1 ● ● ● C2 C1 O1 ● 0.020 ● E1 ● ● L1 J1 0.010 ● I1 0.005 ● V1 0.002 0.001 5 10 15 Median Search Time − One Process (seconds) Figure 28: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and median search time (less than 20 seconds) for a single process for Class A — Left and Right Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 18 and search template creation times can be found in Table 74 in Appendix H. Left and Right Index (Greater Than or Equal To 20−Second Searches) 0 1 2 3 4 5 0.100 ● ● H1 H2 FNIR @ FPIR =10−3 0.050 0.020 F1 ● F2 T2 ● ● ●● ● P1 P2 S1 ● ● U1 K2 ● O2 U2 ● ● G2 ● E2 ● S2 ● J2 0.010 ● L2 0.005 V2 ● ● D1 0.002 0.001 100 Q2 ● D2 ● ● Q1 ● I2 200 300 400 Median Search Time − One Process (seconds) 500 Figure 29: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and median search time (greater than or equal to 20 seconds) for a single process for Class A — Left and Right Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 18 and search template creation times can be found in Table 74 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 46 FPVTE – F INGERPRINT M ATCHING Left Index (Less Than 20−Second Searches) 0 0.20 + + + − FNIR @ FPIR =10−3 5 0.10 − + − − − − − − − − + + − + − − 0.05 − − − 0.02 C1 C2 D1 E1 E2 F1 F2 G1 H1 H2 I1 J1 J2 L1 L2 O1 O2 P2 Q1 Q2 S1 U1 V1 Submission Figure 30: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Left Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. Left Index (Greater Than or Equal To 20−Second Searches) 10 20 30 − FNIR @ FPIR =10−3 0.10 − 0.05 − + − 0.02 D2 G2 I2 S2 V2 Submission Figure 31: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Left Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 47 Right Index (Less Than 20−Second Searches) −5 0 5 + FNIR @ FPIR =10−3 0.20 + + + − 0.10 − − + − − − − − − − + − + − 0.05 − − − − Q1 Q2 − 0.02 C1 C2 D1 E1 E2 F1 F2 G1 H1 H2 I1 J1 J2 L1 L2 O1 O2 P2 S1 T1 U1 V1 Submission Figure 32: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Right Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. Right Index (Greater Than or Equal To 20−Second Searches) 10 20 30 − FNIR @ FPIR =10−3 0.10 − 0.05 − − − 0.02 D2 G2 I2 S2 V2 Submission Figure 33: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Right Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 48 FPVTE – F INGERPRINT M ATCHING Left and Right Index (Less Than 20−Second Searches) −40 −30 −20 −10 0 − FNIR @ FPIR =10−3 0.100 0.050 + + − + 0.020 − 0.010 − − 0.005 C1 C2 G1 I1 J1 L1 O1 V1 Submission Figure 34: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Left and Right Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. Left and Right Index (Greater Than or Equal To 20−Second Searches) 0 100 200 300 − FNIR @ FPIR =10−3 0.200 0.100 − 0.050 − − 0.020 + + + − − − − − + − + 0.010 0.005 − D1 D2 − − − E1 E2 F1 F2 G2 H1 H2 I2 J2 L2 O2 P2 − − Q1 Q2 S1 S2 U1 U2 V2 Submission Figure 35: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class A — Left and Right Index. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D 1 0.26 24 2 6 0.76 25 1 20 7.32 1 Time FNIR 1 0.24 24 0.1337 6 0.69 23 0.1124 0.0197 20 7.46 1 0.0190 0.1335 0.1132 3 0.35 12 0.0745 3 0.32 11 0.0630 2 30 16.90 11 0.0723 29 16.27 10 0.0624 1 11 2.49 20 0.1111 11 2.38 20 0.0933 2 14 3.56 18 0.1082 15 3.60 18 0.0903 1 21 9.74 19 0.1089 24 10.10 19 0.0910 1 15 3.61 26 0.1576 14 3.56 26 0.1230 2 17 4.13 27 0.1607 17 4.05 27 0.1249 1 22 9.94 5 0.0257 22 9.83 3 0.0215 1 4 0.54 14 0.0786 4 0.51 16 0.0708 2 7 1.25 10 0.0712 7 1.13 12 0.0643 1 25 10.44 17 0.0883 26 10.42 14 0.0682 2 23 10.32 16 0.0875 25 10.27 15 0.0685 1 2 0.29 8 0.0625 2 0.26 8 0.0505 2 12 3.32 6 0.0351 13 3.26 6 0.0295 1 10 1.84 29 0.2995 10 1.78 30 0.2615 2 26 10.70 28 0.2921 21 9.39 29 0.2526 1 5 0.62 15 0.0818 5 0.56 17 0.0776 2 9 1.56 13 0.0766 9 1.36 13 0.0675 1 8 1.32 23 0.1308 8 1.24 25 0.1133 2 13 3.33 22 0.1272 12 3.07 22 0.1100 1 28 15.70 2 0.0222 28 14.24 4 0.0218 2 24 10.43 3 0.0226 23 9.98 2 0.0214 1 29 16.86 7 0.0571 30 16.55 7 0.0442 1 16 3.99 30 NA 16 3.73 28 0.1929 2 18 5.97 9 0.0685 19 5.87 9 0.0562 U 1 27 14.42 21 0.1218 27 10.76 21 0.0996 V 1 19 6.01 4 0.0253 18 5.60 5 0.0223 D 2 3 41.84 1 0.0197 2 28.64 1 0.0190 G 2 5 54.03 5 0.1086 5 59.55 5 0.0909 I 2 2 40.99 3 0.0278 3 40.35 2 0.0214 F G H I < 20 seconds Right Index FNIR 1 E J K L M O P Q S T ≥ 20 seconds Left Index Time Sub. # 1 49 S 2 4 44.66 4 0.0650 4 46.05 4 0.0503 U 2 1 24.16 6 0.1178 1 20.10 6 0.1007 V 2 6 65.35 2 0.0252 6 61.24 3 0.0222 Table 17: Tabulation of median identification times for Class A. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 100 000. Time values are median times reported in seconds, but were originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 50 FPVTE – F INGERPRINT M ATCHING Participant Letter < 20 seconds C 1 1 2.08 7 0.0368 2 4 6.35 8 0.0374 G 1 3 5.22 9 0.0515 I 1 8 17.87 2 0.0058 J 1 6 13.64 3 0.0143 K 1 9 18.01 6 0.0360 L 1 2 2.19 4 0.0146 O 1 7 14.46 5 0.0229 V 1 5 9.29 1 0.0034 1 15 70.99 4 0.0030 2 23 237.43 4 0.0030 1 1 16.35 11 0.0207 2 27 518.11 10 0.0202 1 13 60.66 21 0.0386 2 16 73.95 22 0.0412 2 22 221.16 15 0.0311 1 8 36.71 24 0.0686 2 12 52.25 23 0.0684 I 2 25 338.88 4 0.0030 J 2 7 33.35 8 0.0143 K 2 6 32.93 14 0.0286 L 2 3 23.42 7 0.0072 1 5 31.44 26 NA 2 20 171.24 25 NA 2 10 43.53 12 0.0214 1 14 63.65 20 0.0370 2 17 101.16 16 0.0333 1 21 212.69 1 0.0027 2 19 161.02 1 0.0027 D E F G H ≥ 20 seconds Left and Right Index Time FNIR Sub. # M O P Q S T U V 1 2 23.00 13 0.0281 2 26 495.50 9 0.0195 1 4 25.68 27 NA 2 9 37.23 19 0.0366 1 11 45.51 17 0.0336 2 24 240.40 18 0.0358 2 18 127.65 3 0.0028 Table 18: Tabulation of median identification times for Class A. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 1 600 000. Time values are median times reported in seconds, but were originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 8.2 51 Class B Tabulated comparisons of identification times for IDFlat submissions are shown in Table 19. The search times shown in this table are from the “Total/One” column in Tables 30 through 33 included in Appendix C. For reference, the FNIR values from Section 7 are reprinted to the right of the identification times. The tables were used to create scatter plots showing accuracy, search times, and search template creation times. Those plots are shown in Figures 36 through 39. Some observations for Class B identification times include: . Most submissions had some improvement in accuracy with longer search times. . Results vary, but some submissions (D, G, I1, Q1) performed searches faster when more fingers were available, while others (H, L, S) required longer search times with more fingers. . Most search times increased modestly when ten processes were running in parallel, compared to the single process timing test. See Appendix C and Appendix D for complete details. . Tables 30 through 33 in Appendix E and Figures 40 through 43 show differences between the last two rounds of submissions. There are a variety of results in these tables. Again, most had improvement in FNIRs, but some with longer search times and some with dramatically shorter search times. Like Class A, there are certainly indications that high accuracy can be achieved with some fast submissions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 52 FPVTE – F INGERPRINT M ATCHING Left Slap 0 2 4 6 0.2000 8 F1 ● ● M1 C1 0.0500 ● M2 ● H2 U1 ● ● G1 O1 ● ● J1 L2 ● ● E2 V2 ● V1 O2 ● ● J2 D1 ● ● I1 ● 0.0100 U2 ● G2 ● ● L1 ● ● ● S2 ● ● S1 C2 E1 0.0200 14 ● H1 ● ● 12 F2 0.1000 FNIR @ FPIR =10−3 10 D2 Q1 ●● Q2 ● I2 0.0050 0.0020 0.0010 0.0005 20 40 60 Median Search Time − One Process (seconds) 80 Figure 36: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and median search time for a single process for Class B — Left Slap. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 19 and search template creation times can be found in Table 75 in Appendix H. Right Slap 0 2 4 6 8 10 12 14 0.2000 M1 ● ● F1 FNIR @ FPIR =10−3 0.1000 0.0500 ● M2 H1 ● ● C1 0.0200 ● 0.0100 ● F2 ● L1 ● ● H2 S2 ●● S1 C2 ● G1 ● U2 ● G2 ● L2 ● J1 E1 0.0050 ● O1 O2 ● ● J2 V2 ● V1 ● ● E2 D1 ● ● I2 U1 ● I1 ● ● Q2 ● ● Q1 D2 0.0020 0.0010 0.0005 20 40 60 Median Search Time − One Process (seconds) 80 Figure 37: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and median search time for a single process for Class B — Right Slap. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 19 and search template creation times can be found in Table 76 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ● Left and Right Slap ● C1 0 53 C2 5 10 15 20 25 0.2000 FNIR @ FPIR =10−3 0.1000 M1 ●● F1 ● ● F2 M2 0.0500 H1 ● ● H2 S2 ● S1 ●● U2 ● U1 0.0200 ● 0.0100 G2 ● G1 ● E1 0.0050 ● ● ● J1 L2 L1 ● ● E2 ● V2 ● V1 ● 0.0020 O2 ● ● J2 O1 ● I1 Q2 D1 ● D2 ● ● Q1 ● I2 0.0010 0.0005 20 40 60 Median Search Time − One Process (seconds) 80 Figure 38: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and median search time for a single process for Class B — Left and Right Slap. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 19 and search template creation times can be found in Table 77 in Appendix H. IDFlat ● C1 0 5 ● C2 10 15 20 0.2000 FNIR @ FPIR =10−3 0.1000 F1 ● ● ● 0.0500 F2 M1 ● M2 H1 ● 0.0200 U2 ● ● U1 ●S1 0.0100 ● 0.0050 G1 J1 ● G2 ● ● E1 0.0020 O1 ● ● ● L2 L1 ● V1 E2 ● ● ● V2 O2 ● ● J2 D1 ● 0.0010 H2 ● S2 ● I1 Q1 ● ● D2 ● Q2 ● I2 0.0005 20 40 60 Median Search Time − One Process (seconds) 80 Figure 39: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and median search time for a single process for Class B — Identification Flats. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 19 and search template creation times can be found in Table 78 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 54 FPVTE – F INGERPRINT M ATCHING Left Slap 0 20 40 0.20 + FNIR @ FPIR =10−3 + + 0.10 − − − − − − − 0.05 − − − − − − 0.02 − − − − − − 0.01 C1 C2 D1 D2 E1 E2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 M2 O1 O2 Q1 Q2 V1 Submission Figure 40: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class B — Left Slap. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. Right Slap −20 0 20 40 − 0.100 FNIR @ FPIR =10−3 + 0.050 + − + + − − − − − − − − 0.020 − − 0.010 − − − − − − Q1 Q2 0.005 C1 C2 D1 D2 E1 E2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 M2 O1 O2 V1 Submission Figure 41: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class B — Right Slap. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 55 Left and Right Slap −20 0 20 + 0.100 0.050 FNIR @ FPIR =10−3 40 + + − − − − 0.020 − − 0.010 0.005 − − − − − − − − − 0.002 D1 D2 E1 E2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 M2 O1 O2 + + Q1 Q2 V1 Submission Figure 42: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class B — Left and Right Slap. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. Identification Flats −40 −20 0 20 − + FNIR @ FPIR =10−3 0.050 − 0.020 40 − − − − − − 0.010 − − − − − 0.005 − − − 0.002 − + + Q1 Q2 0.001 D1 D2 E1 E2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 M2 O1 O2 V1 Submission Figure 43: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class B — IDFlat. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 56 24 0.0072 0.0392 0.0403 20 5 3 61.24 10.70 6.40 6 29 30 0.0031 NA NA 17 4 3 45.15 10.21 7.92 6 29 30 0.0020 NA NA Identification Flats Time FNIR 3.67 6 23 Left and Right Slap Time FNIR 6.50 FNIR 2 53.32 Right Slap 0.0654 4 Time 22 19 FNIR 3.74 0.0647 Left Slap 2 0.0163 Time 1 6 21 0.0024 6.74 0.0591 52.23 7 0.0591 4 27 0.0062 18 43.61 27 0.0040 2 37.56 18 0.0203 1 15 14 0.0012 12 4.26 49.34 23 0.0043 0.0049 16.26 2 0.0910 1 20 82.66 16 10 0.0901 6 6.76 28 0.0106 26 46.70 26 0.0084 2 33.09 18 0.0349 18 27.27 17 0.0204 0.0024 9 10 3.89 36.36 23 24 0.0012 0.0063 0.0083 34.55 2 5 0.1222 1 12 70.38 86.56 1 15 7 0.1220 11 27 19.07 6.12 29 0.0212 25 7 63.05 34.47 28 0.0198 0.0361 16 0.0049 0.0009 2 31.87 18 0.0641 24 0.0022 17 21 45.92 16 3 43.86 0.0052 8 10 10.13 25 73.06 1 25.38 0.0151 0.0187 37.28 26 37.19 8 2 0.1684 5 15 50.16 13 0.0068 0.0015 13 7 0.1681 13 0.0647 17 16 2.97 29 0.0371 17 26 0.0058 46.13 57.19 34.37 28 0.0325 5 26.15 1 30.09 18 0.0998 52.84 1 7 21 43.59 17 18 55.87 0.0156 0.0045 0.0142 8 10 16.24 23 20 14 0.0259 2 59.49 0.1008 14 41.13 5 1 6 14 49.75 24 0.0116 24.14 13 2 23 4 7 2.82 1 16 51.77 1 0.0287 0.0094 58.10 2 17 55.96 15 1 1 21 38.26 22 2 13 25.61 2 1 7 27 14 0.0057 0.0882 0.0904 0.0062 14 21 11 23 13 9 87.60 66.01 42.92 60.02 35.64 65.91 38.87 28.59 19 21 20 2 2 13 15 25 26 11 0.0027 0.0141 0.0099 0.0136 0.0108 0.0012 0.0012 0.0035 0.0041 0.0515 0.0543 0.0033 1 20.78 25 0.0051 24 88.46 22 0.0024 0.0033 27.15 13 0.0021 29 86.60 9 11 6 45.91 11 0.0022 30 75.28 7 22 8 38.34 2 0.0160 28 34.96 9 0.0202 16 68.46 3 0.0190 25 49.30 24 0.1259 14 65.02 21 0.0139 10 10 17 0.1155 23 53.40 22 0.0124 19 27 30 0.0142 22 83.33 20 0.0036 10 12.66 27 0.0132 19 82.13 19 0.0036 27 32.03 12 0.0057 29 83.62 7 0.0031 6 87.41 11 0.0057 27 82.40 7 10 9 35.61 3 0.0369 30 38.93 60.13 0.0276 29 73.93 3 0.0381 28 52.80 14.50 0.1736 12 64.04 21 0.0266 15 5 14 0.1634 26 61.94 22 0.0273 18 0.0047 30 0.0257 23 70.54 19 0.0106 0.0054 12.37 27 0.0254 22 70.84 20 0.0110 12 31.17 12 0.0098 24 89.26 8 69.28 5 87.21 11 0.0099 25 80.11 9 10.48 1 9 37.01 2 0.1089 30 35.43 4 2 29 73.15 3 0.1133 28 48.51 0.0126 1 12 53.24 25 0.0500 11 0.0167 2 26 54.02 26 0.0461 16 15 1 19 71.69 20 0.0192 5.47 2 20 71.77 19 0.0190 74.09 1 24 80.03 9 3 2 25 89.07 8 0.0236 1 28 36.07 0.0288 2 30 47.97 16 1 11 5.56 2 15 74.38 1 3 2 2 2 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 19: Tabulation of median identification times for Class B. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 3 000 000. Time values are median times reported in seconds, but were originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 8.3 57 Class C Tabulated comparisons of identification times for ten-finger identification submissions are shown in Table 20. The search times shown in this table are from the “Total/One” column in Tables 34 through 36 included in Appendix C. For reference, FNIR values from Section 7 are reprinted to the right of the identification times. The tables were used to create scatter plots showing accuracy, search times, and search template creation times. Those plots are shown in Figure 44. Some observations for Class C identification times include: . Like classes A and B, gains varied across the participants, but most had some level of improvement in accuracy with longer search times. . Results for some submissions varied between ten-finger plain-to-plain and ten-finger rolled-to-rolled impressions. Some were faster with plain impressions and others were faster with rolled impressions. The most accurate submissions appeared to match both types accurately. . Tables 34 through 36 in Appendix E and Figure 45 show differences between the last two rounds of submissions. Like classes A and B, these results indicate that high accuracy can be achieved with some fast submissions, but the absolute best accuracy was not the fastest submission. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 58 FPVTE – F INGERPRINT M ATCHING Plain−to−Plain ● 0 5 C1 10 15 20 25 30 0.2000 ● C2 0.0500 ● M2 ● FNIR @ FPIR =10−3 S2 M1 ● 0.1000 ● ● F1 F2 S1 ● ● G1 H2 ● ● G2 0.0200 ● H1 ● ● U2 0.0100 ● E1 L2 0.0050 ● ● J1 E2 20 ● O1 ● Q2 ●Q1 ● ● I2 0.0010 O2 ● ● V2 ● ● V1 0.0020 0.0005 U1 ● L1 ● ● J2 D1 ● I1 D2 40 60 Median Search Time − One Process (seconds) 80 Rolled−to−Rolled 0 5 10 15 20 25 30 ● S2 0.2000 FNIR @ FPIR =10−3 0.1000 M1 ● ● M2 ● S1 ● 0.0500 ● F1 ● G1 F2 U1 ● ● ● U2 G2 0.0200 H1 ● H2 C1 ● ● 0.0100 E1 ● ● L2 ● C2 L1 0.0050 ● ● J1 E2 ● J2 ● ● O2 O1 0.0020 0.0010 0.0005 20 ● D1 ● ● V1 ● I2 V2 ● I1 Q1●●● D2 Q2 40 60 Median Search Time − One Process (seconds) 80 Plain−to−Rolled 0 5 10 15 20 M1 ● M2 ●● F1 0.2000 25 ● 30 F2 S2 ● S1 ● 0.1000 FNIR @ FPIR =10−3 ● G1 0.0500 ● U1 ● G2 H1 ● ● H2 0.0200 E1 ● L2 ● L1 ● C1 ● E2 ● 0.0050 J1 V1 ● ● Q2 Q1 O1 ● D1 ●● 0.0020 ● I2 20 ● O2 ● ● V2 J2 D2 ● ● 0.0010 0.0005 U2 C2 ● ● 0.0100 ● I1 40 60 Median Search Time − One Process (seconds) 80 Figure 44: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and median search time for a single process for Class C. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot. Median search times are plotted in seconds. The FNIR and median search time data are from Table 20 and search template creation times can be found in Tables 79 and 80 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 59 Ten−Finger Plain−to−Plain −20 0 FNIR @ FPIR =10−3 40 − 0.200 0.100 20 + + − + 0.050 − − − 0.020 − − 0.010 − − − 0.005 − 0.002 − − − − − − 0.001 C2 D1 D2 E1 E2 F1 F2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 Q1 Q2 V1 Submission Ten−Finger Rolled−to−Rolled −60 0.100 − −40 −20 20 40 − + 0.050 FNIR @ FPIR =10−3 0 + + + − − − 0.020 − − − − 0.010 − 0.005 − − − 0.002 + − C1 C2 D1 D2 E1 E2 F1 F2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 Q1 − Q2 − V1 Submission Ten−Finger Plain−to−Rolled −40 −20 + + FNIR @ FPIR =10−3 − 20 − 0.200 0.100 0 − − 0.050 − − − − − 0.020 − − 0.010 − 0.005 − − − + − − − 0.002 0.001 C1 C2 D1 D2 E1 E2 F1 F2 G1 G2 H1 H2 I1 I2 J1 J2 L1 L2 Q1 Q2 V1 Submission Figure 45: Plots showing difference in FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and difference in search times, for a single search process, between round 2 and round 3 submissions for Class C. The “+” symbol indicates that FNIR increased from round 2 to round 3 and “-” indicates a decrease in FNIR. The color of the bar shows the change in search time. The color scale for difference in search time is at the top of the plot and the units are in seconds. The data for the plots are taken from the tables in Appendix E. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 60 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Ten-Finger Plain-to-Plain Time FNIR Sub. # 4 18.30 30 NA 2 5 18.38 24 1 25 73.93 6 2 26 74.36 1 2 12.77 2 17 52.91 1 15 51.69 2 18 1 1 2 1 2 1 Ten-Finger Rolled-to-Rolled Time FNIR 4 20.15 16 0.0094 0.0711 6 21.97 15 0.0015 17 58.92 4 2 0.0011 27 74.97 14 0.0088 2 8.63 13 0.0048 11 42.73 12 25 0.0734 16 55.79 25 60.86 25 0.0734 20 68.20 7.71 23 0.0368 1 7.78 7 23.69 20 0.0276 3 21 68.07 20 0.0276 30 20 65.36 19 0.0275 16 52.41 4 0.0013 2 10 38.28 1 1 12 43.57 12 2 28 77.75 1 3 17.51 2 6 1 11 2 1 Ten-Finger Plain-to-Rolled Time FNIR 6 33.68 17 0.0149 0.0085 8 34.85 18 0.0169 0.0015 13 41.67 6 0.0028 7 0.0018 16 57.36 3 0.0018 18 0.0106 1 8.99 16 0.0137 0.0050 7 34.06 12 0.0056 0.0536 19 58.57 27 0.2514 25 0.0536 21 61.87 27 0.2514 24 0.0447 2 16.78 24 0.0649 19.46 21 0.0333 4 22.14 23 0.0521 84.51 20 0.0201 22 64.83 20 0.0291 29 84.50 19 0.0199 23 65.74 19 0.0285 15 54.56 1 0.0013 15 52.47 2 0.0014 0.0010 7 30.82 2 0.0014 12 41.48 1 0.0011 0.0047 9 33.02 13 0.0051 9 38.41 13 0.0071 10 0.0027 19 67.98 9 0.0033 25 67.88 7 0.0034 16 0.0102 8 31.68 17 0.0097 5 27.47 15 0.0136 23.53 15 0.0095 5 20.25 14 0.0083 3 20.18 14 0.0129 43.37 28 0.0934 12 48.08 28 0.0783 17 57.63 30 0.3067 13 46.70 27 0.0826 13 48.80 27 0.0716 18 57.91 27 0.2514 23 68.54 9 0.0025 14 54.42 11 0.0034 20 59.05 10 0.0041 2 24 73.41 10 0.0027 21 69.30 9 0.0033 24 66.80 8 0.0036 1 9 36.86 2 0.0011 26 74.40 5 0.0017 11 39.76 4 0.0020 2 8 35.35 4 0.0013 25 74.22 2 0.0014 10 39.37 5 0.0022 1 29 86.56 22 0.0311 22 69.71 29 0.0860 30 88.48 25 0.1017 2 30 91.74 29 0.1680 23 70.77 30 0.2462 29 88.07 26 0.2366 1 27 74.94 18 0.0163 28 82.61 23 0.0358 28 82.88 22 0.0378 2 22 68.30 17 0.0155 24 72.48 22 0.0351 27 74.56 21 0.0295 1 14 47.02 7 0.0024 10 40.74 5 0.0017 14 48.40 11 0.0052 2 19 62.26 7 0.0024 18 65.07 8 0.0019 26 69.87 9 0.0039 1 Table 20: Tabulation of median identification times for Class C. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 5 000 000. Time values are median times reported in seconds, but were originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 9 61 Accuracy Computational Resources Tradeoff This section discusses the computational resources used by each submission, mainly looking for trends in accuracy of a submission versus the load it created on the compute nodes. Statistics include how large the stored/finalized templates are on disk versus in memory and how much time it took to create feature templates. Detailed tables in Appendix F show enrollment set sizes, in Appendix G show search template sizes, and in Appendix H show template creation times. All these tables were used to create the scatter plots used in this section. Appendix K plots relative comparisons of FNIR, RAM usage, template creation times, and search times. Addtionally, all the numbers in Tables 60 through 65 in Appendix F are based on the maximum enrollment set sizes for each class shown in Table 4. The RAM values reported are the best estimate based on the information recorded. It is possible that a submission used more or less RAM depending on the internal operations of the submitted software. See the Lessons Learned for Large-Scale Testing section for more details. 9.1 Storage and Memory It is important to note that the Actual RAM used (Appendix F) is the sum of the resident enrollment set sizes of identification stage one processes after returning from the identification stage one initialization method. More information on this can be found in the FpVTE API [15]. Every attempt was made to run each submission on the minimum number of compute nodes needed to successfully complete the evaluation. This generally meant multiple passes of running enrollment set finalization and redoing timing validation tests to determine the minimum number of compute nodes needed. If too few compute nodes were used, the submission would crash and not work properly. When looking at the results, there are submissions like those from participant Q that had large finalized enrollment set, but used a lot less Actual RAM during stage one identification. In fact, Q always ran on a single compute node despite the Finalized storage size. Participant L, on the other hand, clearly compressed templates, so they used more Actual RAM than Finalized storage. This behavior was pre-reported to NIST, which made it easier to plan ahead when testing the submission. 9.1.1 Class A Scatter plots comparing FNIR and computational resources used by index finger identification submissions are shown in Figures 46 through 51. Figures 52 through 53 shows a comparison of the templates (right and left index) as stored on disk with the actual size used in RAM. For the majority of participants, the numbers were very similar but there were exceptions such as T, P, G, L, and Q. Some observations for Class A computational resources include: . The most accurate submissions were Q, V, I, D and L2. . The most accurate submissions used the same or less RAM as other submissions with I2 being an exception. . It appears that high accuracy can be achieved without using a large amount of storage. . Participant T’s submissions consistently used the least computational resources but with the least accuracy. . Of the most accurate submissions, participant V used the least amount of storage space. . Participant K consumed the most RAM and disk space, significantly higher than all other participants. . Participant Q used the least amount of RAM and achieved the highest accuracy of the most accurate submissions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 62 FPVTE – F INGERPRINT M ATCHING Left Index 0 1 2 3 4 5 0.50 FNIR @ FPIR =10−3 ● T1 M1 ● ● M2 0.20 H1 ●● H2 P1 ● C1 ● C2 ● P2 G1 ● ● G2 F1 ● F2 0.10 O1 J1 ●●●● O2 S2 ● T2 ● J2 ● S1 0.05 ● ● U1 U2 K1 ● K2 E1 E2 ● L1 ● ● ● V1 ● V2 Q2 ●● Q1 0.02 I1 ● L2 ● I2 D2 D1 ● 0.01 1 2 RAM Used for Enrollment Set (gigabytes) 3 Figure 46: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and RAM used for enrollment set for Class A — Left Index. The color of the data point is used to show the On Disk Finalized Enrollment Size. The color scale for the On Disk Finalized Enrollment Size is at the top of the plot and the units are in gigabytes. The data for the scatter plot comes from Table 60 in Appendix F. RAM Used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment is from the On Disk/Finalized column. Right Index 0 1 2 3 4 5 0.50 FNIR @ FPIR =10−3 M1 ● ● M2 0.20 ● T1 H2 P1 C1 ● H1 ● ● P2 C2● G1 0.10 F1 ●● ● G2 O1 F2 ● J1 ●●● O2 ● T2 J2 S2 ● 0.05 ● U2 ● U1 K2 ● K1 E1 ● E2 ● L1 ● S1 ● 0.02 0.01 V2 Q1 ●● ● V1 Q2 I1 ● ● D2 D1 L2 ● I2 1 2 RAM Used for Enrollment Set (gigabytes) 3 Figure 47: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and RAM used for enrollment set for Class A — Right Index. The color of the data point is used to show the On Disk Finalized Enrollment Size. The color scale for the On Disk Finalized Enrollment Size is at the top of the plot and the units are in gigabytes. The data for the scatter plot comes from Table 61 in Appendix F. RAM Used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment is from the On Disk/Finalized column. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 63 Left Index 0 2000 4000 6000 8000 10000 0.50 ● T1 FNIR @ FPIR =10−3 M1 ●● M2 0.20 P1 ● ● P2 0.10 H1 ●● H2 C1 ● C2 F1 ●● F2 O1 J1 ●● ●● O2 J2 ● L1 0.05 U2 ● ● G2 U1 ● G1 K1 ● K2 ● E1 ● E2 S2 ● ● S1 ● T2 ● L2 0.02 I1 ● V1 ● V2 ● D1 D2 Q2 ● ● Q1 0.01 2000 ● I2 4000 6000 8000 Search Template Size Per Subject In RAM (bytes) 10000 Figure 48: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and search template size in RAM for Class A — Left Index. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from table Table 66 in Appendix G. Right Index 0 2000 4000 6000 8000 10000 0.50 FNIR @ FPIR =10−3 M1 ●● M2 0.20 0.10 ● T1 P1 ● ● P2 0.05 C1 ● C2 F1 ●● F2 ● H2 ● H1 O1 ● J1 ●● ● O2 J2 L1 U1 ● G2 U2 ● G1 K2 K1 ●● E2 E1 S2 ● ● S1 ● T2 ● 0.02 0.01 V2 ● D2 V1 ● D1 Q2 ● ● Q1 2000 L2 ● I1 4000 6000 8000 Search Template Size Per Subject In RAM (bytes) I2 ● 10000 Figure 49: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and search template size in RAM for Class A — Right Index. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from table Table 66 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 64 FPVTE – F INGERPRINT M ATCHING Left and Right Index 0 10 20 30 40 50 60 70 0.100 H1 ● H2 FNIR @ FPIR =10−3 0.050 G1 ● F2 ● ● ● T2 F1 ● C2 C1 O1 ● S1 O2 J1 ● J2 S2 ● 0.020 P1 ● G2 ● ● ● K1 ● ● U2 P2 ● U1 ● K2 E1 ● ● E2 ● ● L1 0.010 ● L2 ● I1 0.005 Q1 ● Q2 0.002 V1 ● ● V2 0.001 D1 ● D2 ● I2 20 40 RAM Used for Enrollment Set (gigabytes) 60 Figure 50: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and RAM used for enrollment set for Class A — Left and Right Index. The color of the data point is used to show the On Disk Finalized Enrollment Set size. The color scale for the On Disk Finalized Enrollment Set is at the top of the plot and the units are in gigabytes. The data for the enrollment set size comes from Table 62 in Appendix F. RAM used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment Set is from the On Disk/Finalized column. Left and Right Index 0 5000 10000 15000 20000 0.100 H1 ● H2 0.050 P1 ● FNIR @ FPIR =10−3 ● P2 F2 ● F1 ● G1 ● U2 K1 ● ●● U1 ● S1 ●● K2 G2 E1 S2 ● ●● E2 ● C2 C1 O1 ● ● 0.020 ● L1 O2 J1 ● J2 ● T2 0.010 ● L2 0.005 0.002 0.001 Q1 ● Q2 D1 ● D2 ● I1 V1 ● ● V2 5000 10000 15000 Search Template Size Per Subject In RAM (bytes) I2 ● 20000 Figure 51: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and search template size in RAM. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from Table 66 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 65 Left and Right Index L1 32 GB Actual RAM Consumption ● P ● ● E H J ● ●● I2 ● O G C ● ● L2 ● S ● ● I1 ● ● Q ● V ● ● ● U1 U2 K D1 D2 ●● ● M F 1 GB 32 MB T ● 1 GB 4 GB 16 GB 64 GB Finalized Directory Size Figure 52: Comparison of enrollment size in RAM and on disk. The x- and y-axes use log scales. The data for enrollment size comes from Table 62 in Appendix F. Actual RAM Consumption is from the RAM/Actual column and Finalized Directory Size is from the On Disk/Finalized column. Left and Right Index ● 20 kB I2 15 kB RAM Size I1 T ● G ● S E 10 kB ●● U ● ●● ● O D ● ● ● 5 kB C L1 ● ● V L2 K H J ● ● ● P M ● F Q ● ● 0B 0B 5 kB 10 kB 15 kB 20 kB Disk Size Figure 53: Comparison of search template size in RAM and on disk. On Disk comes from Table 66 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 66 FPVTE – F INGERPRINT M ATCHING 9.1.2 Class B Scatter plots comparing FNIR and computational resources used by IDFlat identification submissions are shown in Figures 54 through 55. Figures 56 through 57 shows a comparison of the templates as stored on disk with the actual size used in RAM. Like class A results, this plot highlights submissions where on disk and in RAM usage differed such as E, G, L, and Q. Some observations for Class B computational resources include: . The lowest RAM usage is also one of the top performers (Q). . Other top performers (participants D and I) do not have the largest RAM usage. . Participant I cut RAM usage in half with minimal drop in accuracy. . Like Class A, high accuracy can be achieved while keeping RAM usage relatively low. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING IDFlat ● ● C1 C2 0 67 100 200 300 400 500 600 0.2000 FNIR @ FPIR =10−3 0.1000 M1 ●● M2 0.0500 F2 ● F1 H1 ● H2 0.0200 0.0100 J1 ● ● O1 O2 ●● ● L1 ● V1 J2 V2 ● ● D1 0.0050 E1 ● 0.0020 Q2 ● 0.0010 Q1 ● I1 U2 ● S2 ● ● S1 G1 ● ● U1 G2 ● ● L2 ● E2 ● D2 ● I2 0.0005 100 200 300 400 RAM Used for Enrollment Set (gigabytes) 500 Figure 54: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and RAM used for enrollment set for Class B — Identification Flats. The color of the data point is used to show On Disk Finalized Enrollment Set size. The color scale for the on disk finalized enrollment size is at the top of the plot and the units are in gigabytes. The data for the enrollment set size comes from Table 63 in Appendix F. RAM Used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment Set is from the On Disk/Finalized column. 0 10000 C1 ●● IDFlat C2 20000 30000 40000 50000 60000 70000 80000 0.2000 FNIR @ FPIR =10−3 0.1000 F1 M1 ●●● F2 M2 0.0500 H2 ● H1 0.0200 S2 ● S1 ● 0.0100 0.0050 ● E1 0.0020 0.0010 L2 ● ● L1 V1 ● ● V2 Q2 ● Q1 ● I1 D1 ● J1 ● O1 ●● ● O2 J2 ● U2 ● U1 G1 ● ● G2 ● E2 ● D2 ● I2 0.0005 20000 40000 60000 Search Template Size Per Subject In RAM (bytes) 80000 Figure 55: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and search template size in RAM for Class B — Identification Flats. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from Table 67in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 68 FPVTE – F INGERPRINT M ATCHING Identification Flats ● 512 GB ● U1 U2 ● L2 Actual RAM Consumption 256 GB ● ● G 128 GB E2 S ● H L1 ● E1 64 GB ● ● V I2 ● C 32 GB ● ● F D M ● ● J ● ● ● O I1 16 GB Q 8 GB ● 32 GB 64 GB 128 GB 256 GB 512 GB Finalized Directory Size Figure 56: Comparison of enrollment size in RAM and on disk for Class B — Identification Flats. The x- and y-axes use log scales. The data for enrollment size comes from Table 63 in Appendix F. Actual RAM Consumption is from the RAM/Actual column and Finalized Directory Size is from the On Disk/Finalized column. Identification Flats ● ● 60 kB ● G ● S ● E2 U2 U1 RAM Size 45 kB ● O 30 kB I1 H ● M 15 kB ● E1 ● ● ● ● ●● ● J I2 D V ● F C L Q ● 19 kB 38 kB 57 kB 76 kB Disk Size Figure 57: Comparison of search template size in RAM and on disk for Class B — Identification Flats. On Disk comes from Table 67 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 9.1.3 69 Class C Scatter plots comparing FNIR and computational resources used by plain and rolled impression submissions are shown in Figures 58 through 61. Figures 62 through 65 show a comparison of the templates as stored on disk with the actual size used in RAM. Like class A and B results, these plot highlight submissions where on disk and in RAM usage differed for several submissions. Some observations for Class C computational resources include: . Like classes A and B, the top performers do not use the largest amount of RAM. . High accuracy can be achieved with relatively low RAM usage. . Ten-finger rolled data used more RAM than ten-finger plain data, but is not more accurate. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 70 FPVTE – F INGERPRINT M ATCHING ● 0.2000 200 400 500 600 700 800 900 M1 ● ● M2 F1 ●● C2 F2 0.0500 G1 ● S1 ● ● G2 H1 ● H2 0.0200 0.0100 ● U1 ● ● U2 ● L1 ● E1 L2 J1 ● 0.0050 Q2 ● ● Q1 I1 ● ● ● ● ● E2 O2 ● J2 ● O1 V1 V2 ● 0.0020 0.0010 300 S2 ● 0.1000 FNIR @ FPIR =10−3 Plain C1 100 0 D1 D2 I2 0.0005 200 400 600 RAM Used for Enrollment Set (gigabytes) 800 Figure 58: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and RAM used for enrollment set for Class C — Ten-Finger Rolled-to-Rolled. The color of the data point is used to show the On Disk Finalized Enrollment Set size. The color scale for the On Disk Finalized Enrollment Size is at the top of the plot and the units are in gigabytes. The data for the scatter plot comes from Table 64 in Appendix F. RAM used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment Set is from the On Disk/Finalized column. Rolled 0 100 200 300 400 500 600 700 800 900 ● S2 0.2000 FNIR @ FPIR =10−3 0.1000 M1 ● ● M2 F1 ● F2 0.0500 ● S1 G1 ● U2 ● U1 ● G2 0.0200 H1 ● H2 E1 ● C1 ●● C2 0.0100 L1 ● ● 0.0050 0.0020 Q1 ● ● 0.0010 0.0005 Q2 L2 J1 ● O1 O2 ●● J2 E2 ● V2 ● D2 ● ● D1 ● I2 ● I1 ● V1 200 400 600 RAM Used for Enrollment Set (gigabytes) 800 Figure 59: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and RAM used for enrollment set for Class C — Ten-Finger Rolled-to-Rolled. The color of the data point is used to show the On Disk Finalized Enrollment Set size. The color scale for the On Disk Finalized Enrollment Size is at the top of the plot and the units are in gigabytes. The data for the scatter plot comes from Table 65 in Appendix F. RAM used for Enrollment Set is from the RAM/Actual column and On Disk Finalized Enrollment Set is from the On Disk/Finalized column. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Plain C1 20000 ● 0 40000 60000 100000 120000 40000 60000 80000 100000 Search Template Size Per Subject In RAM (bytes) 120000 0.2000 80000 140000 S2 ● M1 ● M2 ●● F1● F2 C2 0.1000 FNIR @ FPIR =10−3 71 0.0500 G1 ● S1 ● G2 ● U1 ● ● U2 H2 H1 ● 0.0200 L1 ● ● E1 ● L2 0.0100 J1 ● 0.0050 V1 ● V2 0.0020 Q2 ● ● 0.0010 Q1 0.0005 J2 ●● O2 O1 D1 ● ● D2 ● I1 ● E2 ● I2 20000 Figure 60: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and search template size in RAM for Class C — Ten-Finger Plain-to-Plain. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from Table 68 in Appendix G. Rolled 0 20000 40000 60000 80000 0.1000 M1 M2 ●● F1 ● F2 0.0500 0.0200 S1 ● U2 ● U1 L1 ● E1 ● ● L2 ● ● G1 G2 C1 ● ● C2 J1 ● 0.0050 ● O2 O1 J2 ● ● 0.0005 V2 ● Q1 ● ● 0.0010 140000 H1 ● H2 0.0100 0.0020 120000 S2 ● 0.2000 FNIR @ FPIR =10−3 100000 Q2 ● ● ● I1 I2 20000 V1 D1 ● ● E2 D2 40000 60000 80000 100000 Search Template Size Per Subject In RAM (bytes) 120000 Figure 61: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and search template size in RAM for Class C — Ten-Finger Rolled-to-Rolled. The color of the data point is used to show Search Template Size On Disk. The color scale for Search Template Size On Disk is at the top of the plot and the units are in bytes. On Disk comes from Table 68 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 72 FPVTE – F INGERPRINT M ATCHING Ten−Finger Plain Impression ● E2 U ● Actual RAM Consumption 512 GB ● 256 GB S1 ● L2 G ● 128 GB H ● ● L1 ● ● O ● ● E1 C ●● ● V ● F 64 GB J D I1 I2 M ● S2 32 GB Q 16 GB ● 64 GB 128 GB 256 GB 512 GB 1 TB Finalized Directory Size Figure 62: Comparison of enrollment size in RAM and on disk for Class C — Plain Impression. The x- and y-axes use log scales. The data for enrollment size comes from Table 64 in Appendix F. Actual RAM Consumption is from the RAM/Actual column and Finalized Directory Size is from the On Disk/Finalized column. Ten−Finger Plain Impression ● G 64 kB U ● ● E2 ● S RAM Size 48 kB D2 32 kB H I2 I1 C ● 16 kB E1 ● ● ● O ●● ● ● J D1 ● V ● L F ● M Q ● 20 kB 40 kB 60 kB 80 kB Disk Size Figure 63: Comparison of search template size in RAM and on disk for Class C — Plain Impression. On Disk comes from Table 68 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 73 Ten−Finger Rolled Impression 1 TB ● E2 Actual RAM Consumption 512 GB L2 ● ● 128 GB L1 O ● H E1 ● ● F M I1 C ● D ● ● ● ● V G ● ● S1 J ● 256 GB U ● I2 ● S2 64 GB 32 GB Q ● 128 GB 256 GB 512 GB 1 TB Finalized Directory Size Figure 64: Comparison of enrollment size in RAM and on disk for Class C — Rolled Impression. The x- and y-axes use log scales. The data for enrollment size comes from Table 65 in Appendix F. Actual RAM Consumption is from the RAM/Actual column and Finalized Directory Size is from the On Disk/Finalized column. Ten−Finger Rolled Impression ● 108 kB E2 81 kB S ● RAM Size D2 J ● G O 54 kB ● ● C I1 27 kB H E1 ● V U D1 ● ● F ● L ● ● ● ● ● I2 M Q ● 33 kB 66 kB 99 kB 132 kB Disk Size Figure 65: Comparison of search template size in RAM and on disk for Class C — Rolled Impression. On Disk comes from Table 68 in Appendix G. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 74 FPVTE – F INGERPRINT M ATCHING 9.2 Processing Time This section shows the time required to enroll fingerprints images also referred to as template creation time. The template creation times were recorded by enrolling a common sample of the datasets on common hardware, with 100 processes running in parallel across 10 compute nodes (10 processes per compute node). Any segmentation time for slap captures was included in the template creation times. The detailed tables (Tables 72 and 80) used to make the scatter plots (Figures 66 through 71) in this section are included in Appendix H. The Enrollment columns give some idea as to the required system capacity needed to enroll a large dataset in a reasonable time frame. As an example, if a submission takes the full time allowed per image (3 seconds) and 16 compute nodes are used for the enrollment process, it will take approximately 16 days to process all the enrollment sets for all three classes. In an operational sense, enrollment only occurs a single time for the entire dataset, and then on an as-needed basis when new subjects are added to the dataset. The Search columns are different, as they show the time needed to create a template every time a new search is performed. This time would be factored in as part of the overall search time process. Some observations from tables and plots include: . The most accurate submissions did not have the fastest enrollment times. In fact, the best performers tend to have longer enrollment times. . Enrollment time appears to be proportional to finger type and impression. For example, single-finger captures are faster than ten-finger plain impressions, which are faster than ten-finger rolled impressions. . Segmentation does not appear to significantly increase enrollment times, as noted by comparing ten-finger plain impressions to ten-finger rolled impressions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 75 Left Index 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.50 ● FNIR @ FPIR =10−3 T1 0.20 U1 ● ● U2 0.10 ● 0.05 L1 M1 ● ● M2 H2 H1 ●● P1 ● C2 P2 ●●C1 G1 ● G2 O1 ● J1 O2 ●J2 ●●●●● E1 E2 T2 S2 ● K1 ● K2 F1 ● ● F2 ● S1 ● L2 0.02 V1 ● V2 0.01 0.5 ● ● Q2 ● ● Q1 I1 I2 D1 ● D2 1.0 1.5 Enrollment Template Creation Time (seconds) 2.0 Figure 66: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and enrollment template creation time for Class A — Left Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 72 in Appendix H. Right Index 0.0 0.5 1.0 1.5 2.0 2.5 3.0 FNIR @ FPIR =10−3 0.50 M1 ● ● M2 0.20 ● 0.10 0.05 ● L1 T1 H2 ● C2 H1P1 ● ● ● C1 U1 ● P2 ● G2 U2 G1 O1 ● J1 ● O2 ●J2 ● ● E1 E2 ● T2 S2 ● K1 ● K2 F1 ● ● F2 ● S1 ● L2 0.02 0.01 V1 ● V2 Q2 ● ● Q1 0.5 ● I1 ● I2 D1 ● ● D2 1.0 1.5 Enrollment Template Creation Time (seconds) 2.0 Figure 67: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 100 000 subjects and enrollment template creation time for Class A — Right Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 73 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 76 FPVTE – F INGERPRINT M ATCHING Left and Right Index 0 1 2 3 4 5 0.100 FNIR @ FPIR =10−3 0.050 0.020 H1 ● H2 ● G1 C2 U2 ● ● ● C1 T2 ●P1 ● P2 ● ● U1 G2 O1 ● O2 ● ●● E1 E2 ● J1 ● J2 L1 K1 ● S1 ● F2 ● ● F1 ● K2 ● S2 0.010 ● L2 ● I1 0.005 V1 ● Q1 ● Q2 ● V2 0.002 0.001 1 D2 ●● D1 2 3 Enrollment Template Creation Time (seconds) ● I2 4 Figure 68: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 1 600 000 subjects and enrollment template creation time for Class A — Left and Right Index. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 74 in Appendix H. IDFlat ● ● C1 5 C2 0 10 15 20 0.2000 FNIR @ FPIR =10−3 0.1000 F1 F2●●●M1 M2 0.0500 H2 ● H1 0.0200 U2 ● 0.0100 0.0050 ● L2 0.0020 S2 ● ● ● J1 ● ● O1 ● ● O2 ● J2 L1 ● V1 ● V2 S1 U1 ● G1 E1 ● ● G2 E2 ● ● D1 0.0010 Q1 ● Q2 ● D2 ● I1 ● I2 0.0005 5 10 15 Enrollment Template Creation Time (seconds) Figure 69: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 3 000 000 subjects and enrollment template creation time for Class B — Identification Flats. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 78 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Plain ● 0 5 C1 10 0.2000 15 20 25 30 25 30 ● S2 M1 ● ● F1 ●M2 F2 0.1000 ● FNIR @ FPIR =10−3 77 0.0500 C2 G1 ● 0.0200 H1 ● ● G2 H2 U1 ●● U2 L1 ● ● L2 0.0100 0.0050 ● S1 ● E1 ● J1 ● E2 O2 J2 ●●●O1 V2 V1 0.0020 ● D1 0.0010 Q2 ● ● 0.0005 5 I1 ● ● Q1 ● D2 I2 10 15 20 Enrollment Template Creation Time (seconds) Figure 70: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and enrollment template creation time for Class C – Ten-Finger Plain-to-Plain. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 79 in Appendix H. Rolled 0 5 10 15 20 25 30 ● S2 0.2000 0.1000 M1 ● ● M2 ● F1 F2 ● FNIR @ FPIR =10−3 S1 0.0500 G1 ● U2 ● U1 ● G2 H1 ● H2 0.0200 0.0100 L1 ● ● E1 ● L2 J1 ● O1 ● ● O2 J2 0.0050 0.0020 ● C2 ● E2 ● V1 ● V2 Q1 ● ● Q2 0.0010 0.0005 C1 ● 5 ● D1 D2 ● ● I2 ● I1 10 15 20 Enrollment Template Creation Time (seconds) 25 30 Figure 71: Scatter plot of FNIR @ FPIR = 10−3 searching 30 000 subjects against 5 000 000 subjects and enrollment template creation time for Class C – Ten-Finger Rolled-to-Rolled. The color of the data point is used to show the search template creation time. The color scale for search template creation time is at the top of the plot and the units are in seconds. The template creation time data is from Table 80 in Appendix H. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 78 10 FPVTE – F INGERPRINT M ATCHING Ranked Results This section combines tables from Sections 7 and 9 into a single table. The resulting tables are rank-sorted based on FNIR values. There is one table from each class of participation included in the main body of this report. The full set of tables for all classes and search set scenarios are included in Appendix I. These tables are useful because they combine all the high-level information in a single table. They are rank-sorted on FNIR, as accuracy is generally considered the most important goal for an identification algorithm to achieve. The reader can then look across and see how a participant ranked in other areas such as search time (Identification), search template creation time (Search Enrollment), and memory usage (RAM). As stated in previous sections, the most accurate submissions are not the fastest. In all three classes, there is a two to three times increase in the error rate when comparing the most accurate submission with one of the top three fastest in search speed. Appendix K plots relative comparisons of FNIR, RAM usage, template creation times, and search times. Appendix L is one attempt to take the tables in this section and apply some relative weight or importance to each column. The tables in Appendix L use these weights to produce a score for each submission and then sort the results based on those scores. Refer to Appendix L for more details. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Identification Mean Median FNIR 79 Search Enrollment Mean Median RAM Letter Sub. # V 1 1 0.0034 5 9.50 5 9.29 4 0.68 3 0.64 5 11.77 I 1 2 0.0058 9 18.92 8 17.87 9 2.96 9 2.93 6 15.83 J 1 3 0.0143 6 14.00 6 13.64 3 0.66 4 0.65 3 8.30 L 1 4 0.0146 1 2.20 2 2.19 1 0.11 1 0.11 8 18.76 O 1 5 0.0229 7 15.34 7 14.46 2 0.55 2 0.53 4 9.05 K 1 6 0.0360 8 18.32 9 18.01 8 2.11 8 2.09 9 61.81 C 1 7 0.0368 2 2.39 1 2.08 5 0.80 5 0.78 2 4.87 C 2 8 0.0374 4 6.34 4 6.35 5 0.80 5 0.78 1 4.87 G 1 9 0.0515 3 6.27 3 5.22 7 1.13 7 1.09 7 16.38 Q 1 1 0.0027 21 213.08 21 212.69 16 1.13 15 1.08 10 9.14 Q 2 1 0.0027 19 163.65 19 161.02 16 1.13 15 1.08 9 9.14 V 2 3 0.0028 18 133.45 18 127.65 10 0.68 9 0.64 13 11.77 I 2 4 0.0030 25 385.14 25 338.88 27 4.37 27 4.30 21 30.83 D 2 4 0.0030 23 234.52 23 237.43 26 3.12 26 2.84 17 18.58 D 1 4 0.0030 16 73.01 15 70.99 25 3.10 25 2.84 17 18.58 L 2 7 0.0072 3 23.54 3 23.42 3 0.33 3 0.31 26 53.98 J 2 8 0.0143 7 36.19 7 33.35 9 0.66 10 0.65 7 8.30 S 2 9 0.0195 26 429.02 26 495.50 18 1.77 18 1.75 14 14.82 E 2 10 0.0202 27 500.30 27 518.11 11 0.70 11 0.70 23 33.41 E 1 11 0.0207 2 22.27 1 16.35 11 0.70 11 0.70 22 33.40 O 2 12 0.0214 10 45.52 10 43.53 4 0.55 4 0.53 8 9.05 0.0281 1 14.82 S 1 13 20.11 2 23.00 18 1.77 18 1.75 14 K 2 14 0.0286 6 32.68 6 32.93 20 2.11 20 2.09 27 61.81 G 2 15 0.0311 22 227.27 22 221.16 15 1.13 17 1.09 16 16.38 P 2 16 0.0333 17 114.37 17 101.16 13 0.77 13 0.72 19 21.41 U 1 17 0.0336 11 47.60 11 45.51 1 0.30 1 0.30 25 44.63 U 2 18 0.0358 24 252.04 24 240.40 1 0.30 1 0.30 24 37.35 T 2 19 0.0366 9 38.91 9 37.23 5 0.65 7 0.62 1 0.01 P 1 20 0.0370 14 63.41 14 63.65 13 0.77 13 0.72 20 21.42 F 1 21 0.0386 13 59.30 13 60.66 21 2.23 21 2.15 5 4.86 F 2 22 0.0412 15 71.45 16 73.95 21 2.23 21 2.15 5 4.86 H 2 23 0.0684 12 53.77 12 52.25 7 0.66 5 0.60 11 9.36 H 1 24 0.0686 8 37.65 8 36.71 7 0.66 5 0.60 12 9.36 M 2 25 NA 20 183.20 20 171.24 23 2.25 23 2.16 4 3.76 M 1 26 NA 5 32.59 5 31.44 23 2.25 23 2.16 3 3.75 T 1 27 NA 4 26.96 4 25.68 5 0.65 7 0.62 1 0.01 Table 21: Tabulation of ranked results for Class A — Left and Right Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 1 600 000, as seen in Table 18. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 74. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 62. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 80 FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR Search Enrollment Mean Median RAM I 2 1 0.0009 20 60.01 16 43.86 30 19.36 30 19.36 21 108.71 Q 1 2 0.0012 14 48.85 14 42.92 20 8.89 20 8.92 1 7.54 Q 2 2 0.0012 24 71.67 24 66.01 20 8.89 20 8.92 2 7.54 I 1 2 0.0012 6 24.57 7 19.07 25 16.50 25 16.52 5 49.68 D 2 2 0.0012 19 54.37 18 46.70 24 11.90 24 11.86 13 79.43 D 1 6 0.0020 17 52.42 17 45.15 17 6.19 17 6.16 12 79.43 V 2 7 0.0024 15 49.72 19 49.30 3 3.35 7 3.36 9 63.53 E 2 7 0.0024 18 52.88 15 43.61 16 5.95 16 5.93 28 317.95 V 1 9 0.0027 10 35.51 10 34.96 3 3.35 7 3.36 8 63.53 L 1 10 0.0031 5 14.42 5 14.50 2 3.05 2 3.05 22 119.79 L 2 11 0.0033 8 28.56 9 28.59 1 0.88 1 0.88 27 177.48 J 2 11 0.0033 22 64.38 22 60.13 7 3.38 3 3.35 17 101.00 O 2 13 0.0035 21 63.87 21 60.02 5 3.38 5 3.36 20 101.00 G 2 14 0.0040 9 31.67 6 16.26 18 7.89 18 7.90 26 156.12 O 1 15 0.0041 11 37.04 11 35.64 5 3.38 5 3.36 19 101.00 E 1 16 0.0043 2 8.76 2 6.76 13 5.35 13 5.34 14 79.73 J 1 17 0.0049 7 26.31 8 25.38 7 3.38 3 3.35 18 101.00 G 1 18 0.0062 1 6.33 1 4.26 18 7.89 18 7.90 25 156.12 U 1 19 0.0099 30 88.83 28 86.60 15 5.90 15 5.82 29 440.68 S 1 20 0.0108 28 86.50 29 87.60 22 10.24 22 10.31 24 150.35 S 2 21 0.0136 29 88.70 30 88.46 22 10.24 22 10.31 23 150.35 U 2 22 0.0141 25 80.14 25 75.28 14 5.80 14 5.73 30 540.60 H 1 23 0.0203 26 82.43 26 82.66 9 4.37 9 4.37 15 86.46 H 2 24 0.0204 27 85.74 27 86.56 9 4.37 9 4.37 16 86.46 M 2 25 0.0515 23 70.45 23 65.91 26 18.36 28 18.11 6 57.53 M 1 26 0.0543 13 41.37 13 38.87 26 18.36 28 18.11 6 57.53 F 1 27 0.0591 12 39.12 12 37.56 28 18.36 26 18.07 10 77.02 F 2 27 0.0591 16 51.59 20 49.34 28 18.36 26 18.07 10 77.02 C 2 29 NA 4 10.28 4 10.21 11 5.16 11 5.13 4 46.49 C 1 30 NA 3 9.00 3 7.92 11 5.16 11 5.13 3 46.49 Table 22: Tabulation of ranked results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 78. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR 81 Search Enrollment Mean Median RAM I 1 1 0.0013 24 79.04 15 54.56 25 20.21 25 20.23 4 113.80 Q 2 2 0.0014 27 83.35 25 74.22 17 12.95 17 13.06 2 20.22 I 2 2 0.0014 9 40.53 7 30.82 24 18.72 24 18.73 11 137.03 D 1 4 0.0015 17 65.97 17 58.92 23 17.44 21 17.39 10 132.40 Q 1 5 0.0017 26 83.25 26 74.40 17 12.95 17 13.06 1 20.22 V 1 5 0.0017 10 40.94 10 40.74 9 8.48 9 8.50 17 234.03 D 2 7 0.0018 30 86.39 27 74.97 30 30.43 30 30.47 9 132.37 V 2 8 0.0019 16 65.47 18 65.07 9 8.48 9 8.50 18 234.03 O 2 9 0.0033 19 72.55 21 69.30 6 6.80 6 6.71 23 303.13 J 2 9 0.0033 20 72.62 19 67.98 8 6.82 8 6.74 24 303.13 O 1 11 0.0034 15 59.01 14 54.42 6 6.80 6 6.71 25 303.13 E 2 12 0.0050 14 57.02 11 42.73 11 9.89 11 9.91 30 930.48 J 1 13 0.0051 8 36.09 9 33.02 5 6.78 5 6.69 22 303.13 L 2 14 0.0083 3 19.52 5 20.25 3 4.36 3 4.36 26 367.82 C 2 15 0.0085 5 25.91 6 21.97 13 10.79 13 10.69 15 183.36 C 1 16 0.0094 4 25.25 4 20.15 13 10.79 13 10.69 14 183.36 L 1 17 0.0097 7 31.42 8 31.68 3 4.36 3 4.36 21 280.49 E 1 18 0.0106 1 9.52 2 8.63 12 9.90 12 9.92 16 191.66 H 2 19 0.0199 29 84.26 29 84.50 15 12.08 15 12.17 12 144.02 H 1 20 0.0201 28 84.14 30 84.51 15 12.08 15 12.17 13 144.02 G 2 21 0.0333 6 30.61 3 19.46 19 15.55 19 15.50 19 261.29 U 2 22 0.0351 23 74.39 24 72.48 1 2.94 1 2.87 28 806.17 U 1 23 0.0358 25 82.76 28 82.61 1 2.94 1 2.87 28 806.17 G 1 24 0.0447 2 11.67 1 7.78 19 15.55 19 15.50 20 261.29 F 1 25 0.0536 13 55.63 16 55.79 28 21.07 28 20.92 7 130.29 F 2 25 0.0536 18 68.61 20 68.20 28 21.07 28 20.92 6 130.29 M 2 27 0.0716 12 48.71 13 48.80 26 21.02 26 20.89 5 130.29 M 1 28 0.0783 11 47.38 12 48.08 26 21.02 26 20.89 8 130.29 S 1 29 0.0860 22 74.01 22 69.71 21 17.43 22 17.57 27 382.88 S 2 30 0.2462 21 72.95 23 70.77 21 17.43 22 17.57 3 78.28 Table 23: Tabulation of ranked results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 5 000 000, as seen in Table 20. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 80. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 65. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 82 FPVTE – F INGERPRINT M ATCHING 11 How Many Fingers are Needed It is already well known that using more fingers results in a lower FNIR [17]. This section combines the results from Figures 12, 15, 18, and 21 into a single plot in Figure 72. The reader is reminded that enrollment set sizes were 100 000 subjects for single index fingers, 1.6 million subjects for two index fingers, 3 million subjects for IDFlats, and 5 million for ten-finger rolled and plain impressions. The search set size was 30 000 that included 20 000 nonmate and 10 000 mate searches. Additionally, Appendix J plots relative comparisons, by class, for each search set used in FpVTE. Some observations regarding numbers of fingers include: . More fingers were better and produced the most accurate results. . More fingers took more time to enroll (Subsection 9.2) . Ten-finger plain-to-plain impressions were as accurate as ten-finger rolled-to-rolled impressions with higher performing submissions. . More fingers generally produce faster search times against very large enrollment sets (Section 8). . Class B four-finger slap identification appeared to be less accurate than Class A two-finger identification. This needs further investigation as to the cause. Two possibilities are slap segmentation errors or fingerprint image quality. After manually inspecting some of the errors and considering the ten finger IDFlat and plain-to-plain results, it would appear that image quality may have been the largest contributing factor. ● ● 0.200 ● ● ● 0.100 FNIR @ FPIR =10−3 ● ● ● ● 0.050 ● 0.020 ● ● ● ● ● ● ● ● ● 0.010 ● ● ● ● ● ● ● ● ● 0.005 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.002 0.001 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● I2 D2 I1 Q1 Q2 D1 E2 V2 V1 L1 J2 L2 O2 G2 O1 E1 J1 G1 U1 S1 S2 U2 H1 H2 M2 M1 F1 F2 C2 C1 K1 K2 P1 P2 T1 T2 Submission ● Identification Flats ● Left and Right Slap ● Left Slap ● Right Slap ● Ten−Finger Plain−to−Rolled ● Left and Right Index ● Left Index ● Right Index ● Ten−Finger Plain−to−Plain ● Ten−Finger Rolled−to−Rolled Figure 72: Rank-sorted FNIR @ FPIR = 10−3 for All Classes. Submissions “1” and “2” from round 3. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 12 83 FpVTE 2003 Comparison FpVTE 2003 [17] was composed of three separate tests, the Large-Scale Test (LST), the Medium-Scale Test (MST), and the Small-Scale Test (SST). SST and MST tested matching accuracy using individual fingerprints, all of which were images from right index fingers. This contrasts with LST, which evaluated matching accuracy using sets of fingerprint images, where each set includes one to ten finger positions collected from an individual subject at one time. LST used 64 000 fingerprint sets from 25 000 subjects. These fingerprint sets comprised multiple test sets with varying combinations of one, two, four, eight, and ten fingers. MST used 10 000 right index fingers and SST used a subset of 1 000 right index fingers. A significant difference between FpVTE 2012 and FpVTE 2003 testing procedures was that FpVTE 2003 required participants to match all subjects in the datasets against each other and return all 1-to-1 match scores. Therefore, while a direct comparison of results from the two FpVTE evaluations is not possible, this section will look at some of the observations from 2003 and note changes that have occurred in FpVTE 2012. Looking at Figures 73 through 75 (focusing on the “Standard Partition” and “Average TAR”) , a notable observation is that the accuracy gap in 2003 between the most accurate and least accurate systems was very significant. In current results, there is still a measurable accuracy gap, but it doesn’t seem to be nearly as large. In 2003, the accuracy results (as shown in Figure 76) indicated some difficulty measuring the accuracy difference when using four-, eight-, and ten-finger datasets. While it was clear that more fingers produced higher accuracy, it was not clear if ten fingers was significantly better than four fingers. The current FpVTE results used large enough datasets to allow a more accurate measurement of four-, eight-, and ten-finger search sets. There was a noticeable improvement in matching accuracy going from a four-finger search set to a ten-finger search set in FpVTE 2012. Effects of fingerprint quality will be analyzed in a different FpVTE report to see if current technologies have improved when using low quality fingerprint data. Figure 73: FpVTE 2003 SST Results - Range of accuracy on single-finger flats (SST). These systems are sorted by accuracy on the SST standard partition. Note that these results are reported at FAR = 10−3 [17]. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 84 FPVTE – F INGERPRINT M ATCHING Figure 74: FpVTE 2003 MST Results - Range of accuracy across 7 MST partitions. These systems are sorted by accuracy on the standard MST, which is simply the combination of the other partitions [17]. Figure 75: FpVTE 2003 LST Results - Range of accuracy over 27 operational LST partitions. The systems are sorted by their average accuracy over the 27 partitions. Note that sorting by median accuracy would change the order for some systems [17]. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Figure 76: FpVTE 2003 LST Number of Finger Results - Effect of number of fingers and other variables in LST. The y scale is the log of False Reject Rate (FRR), which is 1 − TAR. Note that the single-finger searches (red) are clearly separated from the two-finger searches (green), but the four-, eight-, and ten-finger searches are intermingled. At the test sizes used, accuracy of four-, eight-, and ten-finger searches is difficult to differentiate. The lines off of the top of the chart are for FRR = 0 (perfect results), which cannot be represented in log scale [17]. FPVTE – F INGERPRINT M ATCHING 85 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 86 13 FPVTE – F INGERPRINT M ATCHING Lessons Learned for Large-Scale Testing . Failure Feedback: One of the most difficult aspects of validation was providing useful feedback to participants when failures occurred. The data used in FpVTE was sequestered operational data that could not be shared with participants. Some progress was made in FpVTE by allowing participants to write text-only logs, from their submission executed at NIST, that could then be returned for analysis. NIST reviewed all logs to ensure the logs did not include information related to the imagery or the NIST internal computing environment. Enhancements could be made to the FpVTE API to allow the FpVTE test driver to toggle logging on and off, preventing participants from submitting a separate logging build while maintaining the speed of not logging under normal use. . Two-Stage Matching Data Transfer: The FpVTE API specifies a 4 GB RAM disk to allow submissions to write data during the first stage of identification that could be referred to during the second stage. The intent was for all processes running on a compute node to share the same 4 GB RAM disk. As NIST did not say how many processes would be run in parallel, it created problems for submissions that wrote a large amount of data per process during the first stage of identification. Without increasing the RAM disk size, NIST was forced to run fewer processes on each compute node to avoid overfilling the RAM disk, which increased the evaluation time and wasted compute node resources. For instance, a compute node that would typically run twenty stage one identification processes might be limited to running just two or three. Future evaluations should better define the expected RAM disk usage, a minimum number of concurrent processes, and other requirements needed to safely and effectively run multiple searches in parallel. . Shared Memory: A key feature of the FpVTE API was that a large enrollment set could be loaded in memory and shared among multiple processes in parallel. An important aspect in allowing this shared memory useage was that the memory must remain static after initialization (see Section 5 for details). This caused problems for some participants and took several validation iterations to correct. . Additional Computational Statistics: The FpVTE test driver recorded the resident enrollment set size of identification stage one processes after returning from the identification stage one initialization method. It was expected that participants would use this method to load the entire enrollment set partition into RAM. While this was a fairly good indicator of RAM requirements, some submissions allocated significantly more memory during the core identification stage one method, which in many cases required re-partitioning the enrollment set with an additional compute node. Should FpVTE be repeated, it would be more fair to record additional computational statistics, such as peak RAM consumption over the execution time of the submission, since the RAM usage after initialization did not completely represent the RAM resources required for some submissions to run. . Timing Submissions: Keeping timing fair is a difficult task. The baseline of performing a timing test with only a single FpVTE process running proved most successful at keeping timing fair for all participants. The timing test was run against the full enrollment set with the assumption that using a smaller enrollment set would cause the search times to decrease. In cases where any unusual results were noticed, the timing tests were repeated to verify that the results remained consistent. . Enrollment and Re-enrollment: Enrolling the full datasets (≈ 11.4 million total subjects across all three classes) was not a trivial task. The original assumption was that this enrollment would only be performed for the first submission and not need to be repeated with later submissions. This was not the case and greatly increased the overall time required to complete the evaluation. Any future evaluations of this magnitude should explore performing a “maximum size” template extraction up front, then allowing for adjustments during the finalization stage to only use the minimum amount of information needed by the submissions. This could greatly reduce the need for re-enrollments. Care would need to be taken when reporting the “size” of the enrollment templates for each submission. . Enrollment Size — Disk vs. RAM: Another failed assumption was that reporting the size of the template at extraction would be a good indicator of how much RAM the enrollment set required (i.e., the Actual RAM and Reported RAM columns from the tables Section 9 should be relatively close). For most submissions, this was true. Other submissions either compressed the templates and required more memory than it appeared they would need (at least C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 87 one participant warned NIST that this would be the case) or they required less storage than the template sizes indicated they would need (again, at least one participant sent a warning about this issue). While the FpVTE API tried to prevent this by asking for both memory usage in RAM and on disk, there was confusion among the participants on what values, if any, to return. For example, many participants were confused on how to report memory usage of slap images—the participant would segment the fingers and return a single template, but report back four separate RAM usage values. Future evaluations will need to provide better guidance on this issue. . Consolidations of Nonmate Searches: A large amount of unexpected time was spent performing consolidations on the “back-end” of the searches. There proved to be a lot more consolidations to examine than originally expected. It took two to three months to clean these up before meaningful results could be produced. A significant improvement to this issue was the decision to flip nonmate search images, as discussed in Subsection 6.5. . Operational Sequestered Data: Participants were able to learn things about their specific submissions even though they may not have been one of the top performers. Some participants shared with NIST that they were grateful to test on the large sample of sequestered operational data to which they might not otherwise have access. They may not have had the best performance, but they were looking to learn about limitations with their submissions and make improvements, which was one of stated goals of FpVTE. This should continue to help advance fingerprint matching technologies and support the NIST mission of, “promoting U.S. innovation and industrial competitiveness”. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 88 FPVTE – F INGERPRINT M ATCHING 14 Way Forward NIST plans to publish further research and analysis in addition to this initial results report, including: . running larger search sets (300 000 nonmates and 50 000 mates) so that DET curves can show accuracy with FPIR rates below 10−3 ; . performing failure analysis in an attempt to determine if there are common failures among submissions and what causes those failures. Some things to examine during failure analysis include image quality, segmentation errors, gender differences, and consolidation errors; . performing analysis of results based on NFIQ values for the datasets; . looking at accuracy of subgroups of metadata, such as male versus female. . performing further analysis and testing to determine possible causes for four-finger IDFlat slap images being less accurate than two-index finger single-capture images; If FpVTE were to be repeated, it might be useful to concentrate more on throughput versus accuracy. For instance, rather than set a single maximum search time of 90 seconds, the evaluation could have several search time maximums in an effort to see how different search times impact matching accuracy. It was clear during this evaluation that some participants have finer-grained control over the speed in which searches were performed. A speed-vs-accuracy track/competition would be useful. 14.1 Related Testing 14.1.1 Forensic Palmprint As data becomes available, the protocols from this evaluation could be applied to perform an evaluation for latent palmprint matching. 14.1.2 Mobile Data NIST has performed some testing with simulated mobile data, but future evaluations should look at using operational mobile data in the search sets to see how it impacts performance of matching algorithms. 14.1.3 Cross-Comparison of Modalities Additional testing will compare the performance of fingerprint, face, and iris matching algorithms, in which all use a search set of 1.6 million subjects. While the datasets will be captured from different sources and subjects, this will be one of the first steps in comparing different modalities on similar sample sizes. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 89 References [1] C HAMBERS , J., C LEVELAND , W., K LEINER , B., AND T UKEY, P. Graphical Methods for Data Analysis. Wadsworth Brooks/Cole, 1983. 198 [2] C RIMINAL J USTICE I NFORMATION S ERVICES D IVISION — F EDERAL B UREAU OF I NVESTIGATION. IAFIS-IC-0110 (V3.1) WSQ Gray-scale Fingerprint Image Compression Specification, 2010. 6 [3] G ROTHER , P., M ICHAELS , R., AND P HILLIPS , P. J. Face Recognition Vendor Test 2002 Performance Metrics. In Audioand Video-Based Biometric Person Authentication: 4th International Conference, AVBPA 2003, Guildford, UK, June 9–11, 2003 Proceedings (June 2003), J. Kittler and M. S. Nixon, Eds., vol. 2688, Springer Berlin Heidelberg, pp. 937–945. 1 [4] G ROTHER , P., Q UINN , G., M ATEY, J., N GAN , M., S ALAMON , W., F IUMARA , G., AND WATSON , C. Iris Exchange III. NIST Interagency Report 7836 (2012). 18 [5] I MAGE G ROUP — NIST. evaluations.cfm. 3 NIST Biometric Evaluations Website. http://www.nist.gov/itl/iad/ig/biometric_ [6] I NTERNATIONAL C OMMITTEE FOR I NFORMATION T ECHNOLOGY S TANDARDS. American National Standard for Information Technology - Finger Minutiae Format for Data Interchange, ANSI/INCITS 378-2004, 2004. 3 [7] ISO/IEC JTC 1 SC 37. Information technology – Vocabulary – Part 37: Biometrics. http://standards.iso.org/ ittf/PubliclyAvailableStandards/c055194_ISOIEC_2382-37_2012.zip, 2012. 1 [8] K OMARINSKI , P. Automated Fingerprint Identification Systems (AFIS), 1st ed. Elsevier, 2004. ISBN 9780080475981. 19 [9] M ARTIN , A. F., D ODDINGTON , G. R., T ERRI K AMM , M. O., AND P RZYBOCKI , M. A. The DET curve in assessment of detection task performance, 1997. 18 [10] O PEN MPI P ROJECT. Open MPI: Open Source High Performance Computing. http://www.open-mpi.org/, 2014. [Online; accessed 22 October 2014]. 11 [11] S ALAMON , W., AND F IUMARA , G. Biometric Evaluation Framework. http://www.nist.gov/itl/iad/ig/framework. cfm, 2014. [Online; accessed 28 August 2014]. 11 [12] TABASSI , E., AND W ILSON , C. L. A novel approach to fingerprint image quality. In Image Processing, 2005. ICIP 2005. IEEE International Conference on (Sept 2005), vol. 2, pp. II–37–40. 6 [13] TABASSI , E., W ILSON , C. L., AND WATSON , C. I. Fingerprint Image Quality. NIST Interagency Report 7151 (2004). 6 [14] WATSON , C., G ARRIS , M. D., TABASSI , E., W ILSON , C. L., M C C ABE , R. M., J ANET, S., AND K O , K. User’s Guide to Export Controlled Distribution of NIST Biometric Image Software (NBIS-EC). NIST Interagency Report 7391 (2007). 6 [15] WATSON , C., S ALAMON , W., AND F IUMARA , G. Fingerprint Vendor Technology Evaluation Implementer’s Guide. National Institute of Standards and Technology, http://nigos.nist.gov:8080/evaluations/fpvte2012/FPVTE2012_API_ Plan_full_060712.pdf, July 2012. xvi, 1, 61 [16] WAYMAN , J., J AIN , A., M ALTONI , D., AND M AI , D. Biometric Systems: Technology, Design and Performance Evaluation, 2005. ISBN 1852335963. 19 [17] W ILSON , C., H ICKLIN , R. A., B ONE , M., K ORVES , H., G ROTHER , P., U LERY, B., M ICHEALS , R., Z OEPFL , M., O TTO , S., AND WATSON , C. Fingerprint Vendor Technology Evaluation 2003. NIST Interagency Report 7123 (2004). 82, 83, 84, 85 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 90 A FPVTE – F INGERPRINT M ATCHING Individual Participant FNIR Plots This appendix contains a full size DET curve for every participant in FpVTE. The reader is reminded that enrollment set sizes were 100 000 subjects for single index fingers, 1.6 million subjects for two index fingers, 3 million subjects for IDFlats, and 5 million for ten-finger rolled and plain impressions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 77: DET for Participant C — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 91 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 92 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 78: DET for Participant D — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 79: DET for Participant E — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 93 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 94 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 80: DET for Participant F — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 81: DET for Participant G — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 95 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 96 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 82: DET for Participant H — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 83: DET for Participant I — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 97 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 98 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 84: DET for Participant J — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Right Index (2) Left and Right Index (2) False Positive Identification Rate 0.0100 Left and Right Index (1) Left Index (2) 0.1000 Figure 85: DET for Participant K — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) 0.2000 0.5000 FPVTE – F INGERPRINT M ATCHING 99 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 100 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 86: DET for Participant L — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 87: DET for Participant M — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 101 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 102 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 88: DET for Participant O — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Right Index (2) Left and Right Index (2) False Positive Identification Rate 0.0100 Left and Right Index (1) Left Index (2) 0.1000 Figure 89: DET for Participant P — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) 0.2000 0.5000 FPVTE – F INGERPRINT M ATCHING 103 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 104 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 90: DET for Participant Q — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 91: DET for Participant S — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 105 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 106 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0005 0.0010 Left Index (1) Right Index (1) 0.0020 0.0100 Left and Right Index (1) Left Index (2) 0.0050 0.0200 Right Index (2) Left and Right Index (2) 0.0500 False Positive Identification Rate 0.1000 Figure 92: DET for Participant T — All fingers with maximum enrollment sets for each class. 0.2000 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 0.0005 0.0020 0.0050 0.0200 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) False Positive Identification Rate 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.1000 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) Figure 93: DET for Participant U — All fingers with maximum enrollment sets for each class. 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.5000 FPVTE – F INGERPRINT M ATCHING 107 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 108 False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.0010 Left Index (1) Right Index (1) Left and Right Index (1) Right Slap (1) Left Slap (1) 0.0005 0.0020 0.0100 Left and Right Slap (1) Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) Ten−Finger Plain−to−Rolled (1) 0.0050 0.0500 Left Index (2) Right Index (2) Left and Right Index (2) Right Slap (2) Left Slap (2) 0.0200 False Positive Identification Rate 0.2000 Left and Right Slap (2) Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) Ten−Finger Plain−to−Rolled (2) 0.1000 Figure 94: DET for Participant V — All fingers with maximum enrollment sets for each class. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING B 109 Combined Class DETs and CMCs This appendix contains DET and CMC curves for all classes and participants grouped together. There is one grouping for the participant’s first submission and another for their second submission. The submissions are split for visibility only— “first” and “second” submissions do not imply any sort of logical grouping. The reader is reminded that enrollment set sizes were 100 000 subjects for single index fingers, 1.6 million subjects for two index fingers, 3 million subjects for IDFlats, and 5 million for ten-finger rolled and plain impressions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 110 FPVTE – F INGERPRINT M ATCHING False Negative Identification Rate 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 0.5000 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.0005 P I C Q J D S K E T L F U M G V O H 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 Left Index (1) Right Index (1) Left and Right Index (1) 0.0005 0.0010 0.0020 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.5000 Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) 0.0500 0.1000 0.2000 Right Slap (1) Left Slap (1) Left and Right Slap (1) False Positive Identification Rate 0.0005 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.5000 Ten−Finger Plain−to−Rolled (1) Figure 95: DETs from the first submission for all participants in all classes. 0.5000 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0020 0.0010 0.0005 0.0200 0.0100 0.0050 0.2000 0.1000 0.0500 0.5000 0.0020 0.0010 0.0005 0.0200 0.0100 0.0050 P I 0.0050 0.0100 0.0200 0.2000 0.1000 0.0500 C S K E 0.0500 0.1000 0.2000 0.0005 0.0010 0.0020 False Positive Identification Rate T L F 0.0005 0.0010 0.0020 0.5000 U M G Figure 96: DETs from the second submission for all participants in all classes. Q J D 0.0050 0.0100 0.0200 0.5000 0.5000 0.0500 0.1000 0.2000 0.0005 0.0010 0.0020 V O H Ten−Finger Plain−to−Rolled (2) 0.0050 0.0100 0.0200 Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) 0.0500 0.1000 0.2000 Right Slap (2) Left Slap (2) Left and Right Slap (2) 0.5000 Left Index (2) Right Index (2) Left and Right Index (2) FPVTE – F INGERPRINT M ATCHING 111 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology False Negative Identification Rate FPVTE – F INGERPRINT M ATCHING 112 Miss Rate 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 0.2000 0.1000 0.0500 0.0200 0.0100 0.0050 0.0020 0.0010 E L F M G O H Ten−Finger Plain−to−Rolled (1) D K Identification Flats (1) Ten−Finger Plain−to−Plain (1) Ten−Finger Rolled−to−Rolled (1) C J Right Slap (1) Left Slap (1) Left and Right Slap (1) I 15 V 10 U 5 T 15 S 10 Q 5 P 10 15 Left Index (1) Right Index (1) Left and Right Index (1) 5 Rank Figure 97: CMCs from the first submission for all participants in all classes. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 0.0010 0.0020 0.0050 0.0100 0.0200 0.0500 0.1000 0.2000 5 10 P I C 15 Left Index (2) Right Index (2) Left and Right Index (2) 5 10 S K E 15 Rank T L F Identification Flats (2) Ten−Finger Plain−to−Plain (2) Ten−Finger Rolled−to−Rolled (2) 5 10 U M G 15 V O H Ten−Finger Plain−to−Rolled (2) Figure 98: CMCs from the second submission for all participants in all classes. Q J D Right Slap (2) Left Slap (2) Left and Right Slap (2) FPVTE – F INGERPRINT M ATCHING 113 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Miss Rate 114 C FPVTE – F INGERPRINT M ATCHING Accuracy Time Tradeoff Detailed Tables with Median Values In order to reduce the number of tables in the main body of the report (Section 8), this appendix contains tables that show the search times for each stage of identification, for both a single process and ten processes. The tables in this appendix report median times. For readers interested in mean times, please refer to Appendix D. Class A results are in Tables 24 through 29, Class B results are in Tables 30 through 33, and Class C results are in Tables 34 through 36. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.76 3.89 0.35 16.90 2.49 3.56 9.74 3.61 4.13 9.94 6 17 3 30 12 15 23 16 19 24 1 1 2 1 1 1 1 1 1.56 1.32 3.33 11.25 7.67 16.86 3.99 5.97 1.41 6.01 10 8 14 28 22 29 18 20 9 21 2 1 2 1 1 1 1 1 V 2 2 0.62 1 5 2 1 3.32 13 1 10.70 0.28 2 2 2 10.32 25 1 1.84 10.44 26 2 27 1.25 7 11 0.54 4 2 2 2 0.26 One 1 1 Sub. # U T S Q P O M L K J I H G F E D C Letter 20 7 21 19 30 22 28 13 9 10 5 23 11 15 2 26 27 8 4 25 18 17 24 14 12 29 3 16 6 1 Stage One 5.81 1.14 7.15 5.43 17.66 7.73 11.61 3.20 1.36 1.62 0.61 9.58 1.78 3.79 0.27 11.18 11.21 1.28 0.56 10.36 5.09 4.54 10.10 3.52 2.40 16.93 0.38 3.90 0.72 0.26 Ten 1 30 1 1 1 27 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 1 1 — 12.96 — — — 2.64 4.48 — — — — — — — — — — — — — — — — — — — — 3.54 — — One 1 30 1 1 1 28 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 27 1 1 Stage Two — — — 17.63 — — — 1.70 1.79 — — — — — — — — — — — — — — — — — — — — 1.62 Ten 19 27 18 16 29 24 28 13 8 9 5 26 10 12 2 23 25 7 4 22 17 15 21 14 11 30 3 20 6 1 6.01 14.42 5.97 3.99 16.86 10.43 15.70 3.33 1.32 1.56 0.62 10.70 1.84 3.32 0.29 10.32 10.44 1.25 0.54 9.94 4.13 3.61 9.74 3.56 2.49 16.90 0.35 7.32 0.76 0.26 One 19 30 20 17 29 21 27 12 8 9 5 22 10 14 2 25 26 7 4 24 16 15 23 13 11 28 3 18 6 1 Total 5.81 18.76 7.15 5.44 17.66 9.45 13.44 3.20 1.36 1.62 0.61 9.58 1.78 3.79 0.27 11.18 11.21 1.28 0.56 10.36 5.09 4.54 10.10 3.52 2.40 16.93 0.38 5.56 0.72 0.26 Ten 0.0226 0.0571 NA 0.0685 0.1218 0.0253 7 30 9 21 4 0.0222 3 0.1272 2 0.1308 23 22 0.0766 0.0818 15 13 0.2921 0.2995 28 0.0351 6 0.0625 8 29 0.0875 0.0883 17 16 0.0712 0.0786 10 0.0257 5 14 0.1576 26 0.1607 0.1089 19 27 0.1082 0.1111 18 0.0723 0.0745 12 20 0.0197 1 11 0.1337 0.1335 25 24 FNIR @ FPIR = 10−3 Table 24: Tabulation of median identification time results for Class A — Left Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. < 20 seconds Participant FPVTE – F INGERPRINT M ATCHING 115 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 116 6 — 22.44 1 1 6 — — — 3.14 1 4 2 5 3 65.35 24.16 44.66 40.99 54.03 41.84 6 2 4 3 5 1 64.02 39.70 47.82 43.35 56.40 21.60 2 6 4 3 5 1 0.0252 0.1178 0.0650 0.0278 0.1086 0.0197 FNIR @ FPIR = 10−3 18.37 1 — 1 0.35 6 Total 1 56.40 1 — 5 — Stage Two 18.19 5 43.35 1 0.41 1 Stage One Sub. # 1 54.03 3 47.82 5 — Participant Letter 2 5 40.99 4 39.38 1 Ten D 2 3 44.66 2 64.02 One G 2 4 23.74 6 Ten I 2 2 65.35 One S 2 6 Ten U 2 One V Table 25: Tabulation of median identification time results for Class A — Left Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. ≥ 20 seconds C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.69 3.77 0.32 16.27 2.38 3.60 10.10 3.56 4.05 9.83 6 18 3 29 12 16 25 15 19 24 1 1 2 1 1 1 1 1 1.36 1.24 3.07 10.94 7.39 16.55 3.73 5.86 1.24 5.60 10 9 13 28 22 30 17 21 8 20 2 1 2 1 1 1 1 1 V 2 2 0.56 1 5 2 1 3.26 14 1 9.39 0.26 2 2 2 10.26 26 1 1.78 10.42 27 2 23 1.13 7 11 0.51 4 2 2 2 0.24 One 1 1 Sub. # U T S Q P O M L K J I H G F E D C Letter 19 7 21 20 30 22 28 13 9 10 5 23 11 16 2 27 26 8 4 24 18 17 25 14 12 29 3 15 6 1 Stage One 5.61 1.01 7.21 5.66 16.75 7.57 11.33 3.20 1.28 1.53 0.58 8.24 1.64 3.71 0.27 10.91 10.81 1.20 0.53 9.86 5.05 4.58 10.23 3.36 2.34 16.58 0.35 3.68 0.68 0.24 Ten 1 30 1 1 1 27 28 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 29 1 1 — 9.19 — — — 2.53 3.24 — — — — — — — — — — — — — — — — — — — — 3.69 — — One 1 30 1 1 1 28 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 27 1 1 Stage Two — — — 14.92 — — — 1.67 1.73 — — — — — — — — — — — — — — — — — — — — 1.56 Ten 18 27 19 16 30 23 28 12 8 9 5 21 10 13 2 25 26 7 4 22 17 14 24 15 11 29 3 20 6 1 5.60 10.76 5.87 3.73 16.55 9.98 14.24 3.07 1.24 1.36 0.56 9.39 1.78 3.26 0.26 10.27 10.42 1.13 0.51 9.83 4.05 3.56 10.10 3.60 2.38 16.27 0.32 7.46 0.69 0.24 One 18 28 20 19 30 22 27 12 8 9 5 21 10 14 2 26 25 7 4 23 16 15 24 13 11 29 3 17 6 1 Total 5.61 15.85 7.21 5.66 16.75 9.24 13.09 3.20 1.28 1.53 0.58 8.24 1.64 3.71 0.27 10.91 10.81 1.20 0.53 9.86 5.06 4.59 10.23 3.36 2.34 16.58 0.35 5.27 0.68 0.24 Ten 0.0996 0.0223 21 5 0.1929 28 0.0562 0.0442 7 9 0.0214 0.0218 2 0.1100 4 0.1133 25 22 0.0675 0.0776 17 13 0.2526 0.2615 29 0.0295 6 0.0505 8 30 0.0685 0.0682 14 15 0.0643 0.0708 12 0.0215 3 16 0.1230 26 0.1249 0.0910 19 27 0.0903 0.0933 18 0.0624 0.0630 11 20 0.0190 1 10 0.1124 0.1132 23 24 FNIR @ FPIR = 10−3 Table 26: Tabulation of median identification time results for Class A — Right Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. < 20 seconds Participant FPVTE – F INGERPRINT M ATCHING 117 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 118 6 — 11.53 1 1 6 — — — 3.02 1 4 3 5 2 61.24 20.10 46.05 40.35 59.55 28.64 6 2 4 3 5 1 60.53 34.54 45.80 42.08 58.30 20.55 3 6 4 2 5 1 0.0222 0.1007 0.0503 0.0214 0.0909 0.0190 FNIR @ FPIR = 10−3 17.53 1 — 1 0.32 6 Total 1 58.30 1 — 5 — Stage Two 17.43 5 42.08 1 0.33 1 Stage One Sub. # 1 59.55 3 45.80 5 — Participant Letter 2 5 40.35 4 34.25 1 Ten D 2 3 46.05 2 60.53 One G 2 4 19.74 6 Ten I 2 2 61.24 One S 2 6 Ten U 2 One V Table 27: Tabulation of median identification time results for Class A — Right Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. ≥ 20 seconds C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology < 20 seconds 1 1 1 1 1 1 I J K L O V 2.08 6.35 5.22 17.87 13.64 18.01 2.19 14.46 9.29 1 4 3 8 6 9 2 7 5 One 5 7 2 9 6 8 4 3 1 Stage One 9.45 15.50 5.05 18.95 14.42 18.87 6.62 6.54 2.23 Ten 1 1 1 1 1 1 1 1 1 — — — — — — — — — One 1 1 1 1 1 1 1 1 1 Stage Two — — — — — — — — — Ten 5 7 2 9 6 8 3 4 1 9.29 14.46 2.19 18.01 13.64 17.87 5.22 6.35 2.08 One 5 7 2 9 6 8 4 3 1 Total 9.45 15.50 5.05 18.95 14.42 18.87 6.62 6.54 2.23 Ten 0.0374 0.0515 0.0058 0.0143 0.0360 0.0146 0.0229 0.0034 9 2 3 6 4 5 1 0.0368 8 7 FNIR @ FPIR = 10−3 Table 28: Tabulation of median identification time results for Class A — Left and Right Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. 1 2 1 Sub. # G C Letter Participant FPVTE – F INGERPRINT M ATCHING C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 119 FPVTE – F INGERPRINT M ATCHING 120 15 237.43 70.99 27 1 21 6 18.25 182.69 37.59 10 11 4 4 0.0202 0.0207 0.0030 0.0030 FNIR @ FPIR = 10−3 23 16.35 Ten 19.23 1 Total 24 7.29 27 One 52.34 23 — Ten 24 68.09 1 Stage Two 19.30 26 1 One 2 169.73 — Ten 19.27 21 1 Stage One 2 164.19 1 One 1 21 18.25 Sub. # 2 1 0.0412 27 22 0.0311 16.35 15 0.0686 1 68.04 24 0.0684 27 13 233.71 23 0.0030 1 22 39.83 4 H M C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Participant Letter D E 73.95 7 54.92 0.0386 16 221.16 10 21 22 36.71 24 57.10 — 8 52.25 0.0143 512.28 1 — 12 8 11 1 — 25 36.28 377.31 0.0286 60.66 — 1 — 5 14 0.0072 518.11 1 — 1 33.35 338.88 7 NA 13 1 — 1 7 33.62 26 NA — 68.04 1 — — 4 69.95 25 0.0214 — 13 233.70 1 — 32.93 14 30.23 12 1 23 39.83 1 1 6 23.42 3 169.90 — 73.95 8 54.92 — 3 31.44 8 19 — 16 221.16 10 — — 5 171.24 1 1 23 36.71 24 1 1 — 20 57.10 2 10 52.25 36.28 377.30 — 1 — 10 512.28 2 13 7 1 — 1 — 11 1 25 33.35 338.88 33.62 1 — 1 60.66 2 9 6 69.95 1 1 518.11 2 32.93 14 30.23 — 14 2 8 23.42 5 1 2 J I 2 4 31.44 1 F K 2 7 169.90 G L 1 9 22 0.0333 171.24 16 0.0027 22 1 0.0027 12 105.73 1 0.0281 2 16 242.28 13 0.0195 0.0370 23 175.09 9 20 101.16 20 22.50 46.77 17 212.69 2 62.81 21 161.02 26 12 — 19 23.00 43.53 1 80.55 2 63.65 26 83.71 26 14 — 27 — — 1 57.36 1 — 25 69.98 1 1 105.73 27 — — 17 160.28 1 — 20 89.06 1 1 101.16 16 22.50 46.77 18 153.61 3 62.81 1 20 86.83 26 12 2 17 23.00 43.53 1 3 NA 63.65 2 26 27 15 1 88.13 0.0366 487.65 19 0.0336 15 131.08 17 0.0358 25.68 18 52.79 18 0.0028 495.50 37.23 9 407.99 3 4 9 45.51 25 128.23 — — 11 240.40 17 — 1 25.27 24 127.65 1 — 25 0.63 18 — 1 17.03 22 — — 131.01 23 0.68 1 1 19 27.53 22 — 88.07 37.23 4 407.47 1 487.65 1 11 28.49 25 128.23 15 2 6 239.62 18 25.68 1 24 127.65 495.50 2 19 5 2 2 2 O P Q S T U V Table 29: Tabulation of median identification time results for Class A — Left and Right Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. ≥ 20 seconds 50.78 71.69 71.77 33.57 42.79 36.07 47.97 21 26 27 10 15 12 18 2 1 2 1 2 1 2 50.00 20 1 1 73.15 2 28 74.38 29 1 2 25.61 7 2 37.01 38.26 14 1 13 55.96 24 2 1 51.77 22 1 87.21 49.75 19 2 30 59.49 25 1 2 16.24 6 2 31.17 43.59 16 1 9 30.09 8 2 1 34.37 11 1 5.56 2.82 1 12.37 52.89 23 2 5 47.93 17 1 2 6.74 4 2 3 3.74 One 2 1 Sub. # NA — 15 50.36 36.78 NA — 11 74.84 74.67 53.59 54.90 77.48 39.34 92.65 33.44 30.25 10.01 78.75 27.09 42.90 66.74 60.07 61.05 60.00 17.59 47.82 32.85 35.02 3.19 80.04 70.13 7.03 3.88 Ten 24 23 16 17 25 12 28 9 7 4 26 6 13 21 19 20 18 5 14 8 10 1 27 22 3 2 Stage One 1 1 29 30 1 1 27 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 25 1 1 — — 45.59 47.07 — — 3.55 3.49 — — — — — — — — — — — — — — — — — — 4.05 3.29 — — One 1 1 — — 1 1 26 25 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 3.32 3.20 — — — — — — — — — — — — — — — — — — 9.04 6.08 Ten 15 11 30 28 25 24 20 19 26 12 29 9 5 3 27 7 13 21 17 16 23 6 14 8 10 1 22 18 4 2 47.97 36.07 89.07 80.03 71.77 71.69 54.02 53.24 73.15 37.01 87.21 31.17 12.37 5.56 74.38 25.61 38.26 55.96 51.77 49.75 59.49 16.24 43.59 30.09 34.37 2.82 58.10 52.23 6.74 3.74 One 15 11 — — 23 22 16 17 25 12 28 9 7 4 26 6 13 21 19 20 18 5 14 8 10 1 27 24 3 2 Total 50.37 36.78 NA NA 74.84 74.67 57.02 58.21 77.48 39.34 92.65 33.44 30.25 10.01 78.75 27.09 42.90 66.74 60.07 61.05 60.00 17.59 47.82 32.85 35.02 3.19 89.04 76.64 7.03 3.88 Ten 0.0654 0.0647 0.0163 0.0142 0.0259 0.0187 0.1684 0.1681 0.0371 0.0325 0.0998 0.1008 0.0116 0.0094 0.0287 0.0236 0.0288 0.0276 0.1736 0.1634 0.0257 0.0254 0.0098 0.0099 0.1089 0.1133 0.0500 0.0461 0.0192 0.0190 22 21 6 5 13 7 29 28 18 17 23 24 4 1 15 10 16 14 30 27 12 11 2 3 25 26 20 19 9 8 FNIR @ FPIR = 10−3 Table 30: Tabulation of median identification time results for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 10 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 121 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 122 — 4 2 53.32 6.50 3.67 24 3 2 76.16 6.78 3.75 13 2 6 23 24 0.0151 0.0052 0.0072 0.0392 0.0403 FNIR @ FPIR = 10−3 1 — 19 92.10 Total — 1 5.79 1 27 Stage Two 1 — 27 57.19 Stage One 3.74 1 3.56 1 21 Ten 2 6.78 26 8.74 One 3.67 3 69.48 1 28 Ten 2 6.50 22 4.41 One 1 4 48.43 1 28 Ten 2 18 82.56 One 1 1 27 0.1222 52.47 29 0.1220 1 28 0.0212 21 33.38 18 0.0198 2 8 48.57 16 0.0641 0.0083 15 10.56 25 7 31.87 4 39.25 3.23 8 45.92 12 34.32 15 10.13 18 10 — 5 37.28 2.97 1 — 13 34.47 1 — 17 10 — 1 — — 1 — 1 — 1 — 1 1 33.38 1 — — 8 48.57 1 — 15 10.56 1 1 31.87 4 39.25 3.23 8 45.92 12 34.32 2 17 10.13 18 10 1 5 37.28 2.97 2 14 0.0647 34.47 1 20 26 11 2 61.42 0.0058 60.56 5 0.0045 17 66.70 1 0.0156 50.16 20 48.17 14 0.0126 52.84 55.87 14 26.83 10 18 20 41.13 6 80.24 — — 14 24.14 26 — 1 — 7 74.09 1 — 1 — 27 — 1 — 1 — — 66.70 1 — 1 0.0167 1 21 48.17 1 — 15 61.42 55.87 14 26.83 1 5 60.56 2 23 41.13 6 80.24 3 17 1 16 24.14 26 1 50.16 2 7 74.08 1 0.0202 52.84 1 29 5 17 22 2 3 10.64 0.1259 31.81 30 0.1155 7 34.25 27 0.0142 5.47 9 92.89 12 0.0132 12.66 32.03 28 39.42 11 6 9 87.41 13 78.87 — — 29 35.61 25 — 1 — 12 73.93 1 — 1 — 26 — 1 — 1 — — 34.25 1 — 1 0.0057 1 9 92.89 1 — 3 10.64 32.03 28 39.42 1 21 31.80 2 9 87.41 13 78.87 23 7 1 30 35.61 25 25 5.47 2 13 73.93 27 12.66 1 28 20 6 2 25 0.0369 0.0057 21 0.0381 3 22 0.0266 69.69 74.64 19 0.0273 66.61 22 75.02 20 0.0106 19 70.54 23 NA 8 0.0110 64.04 24 70.84 — NA 9 61.94 — 25 89.26 — 37.08 22 1 — 30 80.11 11 50.24 3.25 — 1 NA 28 35.43 16 3.30 1 — — NA 11 48.51 26 74.64 1 54.61 — — 16 3.65 23 75.02 30 40.70 1 — 3.54 70.54 24 NA 29 — 1 25 2 26 70.84 — NA 1 — 66.42 1 27 34.38 — 37.08 1 63.34 2 10 40.66 11 50.24 19 1 15 35.43 16 60.64 2 12 48.51 58.39 1 19 24 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 31: Tabulation of median identification time results for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 11 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 52.39 83.33 82.13 29.82 38.83 38.93 52.79 20 30 29 10 16 17 21 2 1 2 1 2 1 2 63.79 24 1 1 68.46 2 25 69.28 26 1 2 26.15 7 2 38.34 46.13 19 1 15 37.19 14 2 1 73.06 28 1 45.91 70.38 27 2 18 34.55 12 1 2 3.89 1 2 27.15 36.36 13 1 8 27.27 9 2 1 33.09 11 1 20.78 6.12 2 10.47 60.79 23 2 6 59.32 22 1 2 10.70 5 2 4 6.40 One 3 1 Sub. # NA — 17 55.29 41.09 NA — 12 85.11 85.38 57.81 69.86 73.17 41.51 51.07 30.00 63.81 22.48 74.42 28.45 54.07 42.38 85.16 86.18 34.99 4.17 40.13 30.15 35.61 6.09 106.49 104.95 11.10 6.63 Ten 23 25 18 20 21 13 15 7 19 5 22 6 16 14 24 26 9 1 11 8 10 2 28 27 4 3 Stage One 1 1 29 30 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 — — 43.88 53.54 — — 1.05 1.07 — — — — — — — — — — — — — — — — — — 2.28 1.34 — — One 1 1 — — 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 0.76 0.79 — — — — — — — — — — — — — — — — — — 4.43 2.40 Ten 18 15 28 30 27 29 19 22 23 14 16 8 6 4 24 7 17 13 26 25 11 1 12 9 10 2 21 20 5 3 52.80 38.93 82.40 83.62 82.13 83.33 53.40 65.02 68.46 38.34 45.91 27.15 20.78 10.48 69.28 26.15 46.13 37.19 73.06 70.38 34.55 3.89 36.36 27.27 33.09 6.12 63.05 61.24 10.70 6.40 One 17 12 — — 23 25 18 20 21 13 15 7 19 5 22 6 16 14 24 26 9 1 11 8 10 2 28 27 4 3 Total 55.29 41.09 NA NA 85.11 85.38 58.57 70.73 73.17 41.51 51.07 30.01 63.82 22.48 74.42 28.45 54.07 42.38 85.16 86.18 34.99 4.17 40.13 30.15 35.61 6.09 111.35 108.01 11.10 6.63 Ten NA NA 0.0031 0.0024 0.0063 0.0049 0.0910 0.0901 0.0106 0.0084 0.0349 0.0361 0.0022 0.0015 0.0068 0.0047 0.0054 0.0062 0.0904 0.0882 0.0057 0.0051 0.0021 0.0022 0.0160 0.0190 0.0139 0.0124 0.0036 0.0036 30 29 6 5 15 10 28 26 18 17 23 24 3 1 16 9 12 14 27 25 13 11 2 3 21 22 20 19 7 7 FNIR @ FPIR = 10−3 Table 32: Tabulation of median identification time results for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 12 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 123 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 124 — 4 3 45.15 10.21 7.92 22 4 3 80.68 10.62 7.99 16 2 6 29 30 0.0043 0.0012 0.0020 NA NA FNIR @ FPIR = 10−3 1 — 17 86.69 Total — 1 1.33 2 24 Stage Two 1 — 27 46.70 Stage One 7.99 1 0.86 2 18 Ten 3 10.62 25 2.30 One 7.92 4 79.06 1 28 Ten 3 10.21 22 1.43 One 1 4 44.48 1 26 Ten 2 19 83.76 One 1 2 23 0.0591 45.79 27 0.0591 2 27 0.0062 20 40.93 18 0.0040 2 11 54.18 14 0.0203 0.0024 17 4.52 23 7 37.56 1 14.99 7.24 12 49.34 5 48.02 20 4.26 27 13 — 1 16.26 6.76 1 — 6 43.61 1 — 26 15 — 1 — — 1 — 1 — 1 — 1 1 40.93 1 — — 11 54.18 1 — 17 4.52 1 1 37.56 1 14.99 7.23 13 49.34 5 48.02 2 22 4.26 27 13 1 1 16.26 6.76 2 6 0.0204 43.61 1 27 24 17 2 97.33 0.0012 98.56 2 0.0009 28 21.52 1 0.0049 82.66 6 54.10 17 0.0033 86.56 19.07 16 26.88 11 27 7 43.86 7 64.19 — — 16 25.38 19 — 1 — 8 60.13 1 — 1 — 22 — 1 — 1 — — 21.52 1 — 1 0.0031 1 6 54.10 1 — 10 97.33 19.07 16 26.88 1 8 98.56 2 7 43.86 7 64.19 5 28 1 18 25.38 19 1 82.66 2 8 60.13 1 0.0033 86.56 1 24 8 11 28 2 5 34.39 0.0543 86.02 26 0.0515 23 42.78 25 0.0041 14.50 12 72.75 15 0.0035 28.59 38.87 20 38.07 13 9 13 65.91 10 63.40 — — 23 35.64 18 — 1 — 11 60.02 1 — 1 — 21 — 1 — 1 — — 42.78 1 — 1 0.0012 1 12 72.75 1 — 2 34.38 38.87 20 38.07 1 14 86.01 2 14 65.91 10 63.40 14 24 1 26 35.64 18 25 14.49 2 12 60.02 28 28.59 1 23 14 9 2 16 0.0108 0.0012 20 0.0136 2 21 0.0099 50.75 90.62 19 0.0141 79.70 25 92.93 22 0.0027 21 87.60 26 NA 9 0.0024 42.92 29 88.46 — NA 7 66.01 — 30 86.60 — 37.51 24 1 — 28 75.28 9 51.13 1.00 — 1 NA 25 34.96 15 1.02 1 — — NA 10 49.30 26 90.62 1 57.06 — — 19 1.48 25 92.93 30 35.79 1 — 1.48 87.60 26 NA 29 — 1 27 2 29 88.46 — NA 1 — 49.75 1 30 29.91 — 37.51 1 78.68 2 10 41.20 9 51.13 21 1 15 34.96 15 41.50 2 11 49.30 64.62 1 21 25 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 33: Tabulation of median identification time results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 13 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 33.71 86.56 91.74 47.96 51.07 47.02 62.26 8 29 30 15 16 14 21 2 1 2 1 2 1 2 35.17 9 1 1 73.41 2 27 77.75 28 1 2 43.57 12 2 68.54 38.28 10 1 24 52.41 18 2 1 65.36 22 1 46.70 68.07 2 13 23.69 7 23 1 2 7.71 1 2 43.37 60.86 20 1 11 51.69 17 2 1 52.91 19 1 20.73 12.77 2 17.50 71.36 25 2 6 72.26 26 1 2 18.38 5 2 3 18.30 One 4 1 Sub. # NA — 18 65.76 48.95 NA — 13 98.32 95.71 35.06 38.39 83.44 74.17 46.86 43.13 66.62 39.57 84.93 47.12 40.63 63.47 73.09 78.40 30.21 8.79 61.46 51.98 54.82 13.68 137.48 141.45 19.05 19.19 Ten 26 25 6 7 23 21 11 10 19 8 24 12 9 17 20 22 5 1 16 14 15 2 27 28 3 4 Stage One 1 1 29 30 1 1 28 27 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 26 25 1 1 — — 17.32 26.89 — — 1.70 1.63 — — — — — — — — — — — — — — — — — — 1.63 1.36 — — One 1 1 — — 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 1.36 1.51 — — — — — — — — — — — — — — — — — — 2.10 1.85 Ten 19 14 22 27 30 29 8 9 24 23 13 11 6 3 28 12 10 16 20 21 7 1 18 15 17 2 26 25 5 4 62.26 47.02 68.30 74.94 91.74 86.56 35.35 36.86 73.41 68.54 46.70 43.37 23.53 17.51 77.75 43.57 38.28 52.41 65.36 68.07 23.69 7.71 60.86 51.69 52.91 12.77 74.36 73.93 18.38 18.30 One 18 13 — — 26 25 6 8 23 20 11 10 21 7 24 12 9 17 19 22 5 1 16 14 15 2 27 28 3 4 Total 65.76 48.95 NA NA 98.32 95.71 36.23 39.96 83.44 74.17 46.86 43.13 74.43 39.58 84.93 47.12 40.63 63.47 73.10 78.40 30.21 8.79 61.46 51.98 54.82 13.68 140.57 143.53 19.05 19.19 Ten NA 0.0711 0.0015 0.0011 0.0088 0.0048 0.0734 0.0734 0.0368 0.0276 0.0276 0.0275 0.0013 0.0010 0.0047 0.0027 0.0102 0.0095 0.0934 0.0826 0.0025 0.0027 0.0011 0.0013 0.0311 0.1680 0.0163 0.0155 0.0024 0.0024 30 24 6 2 14 13 25 25 23 20 20 19 4 1 12 10 16 15 28 27 9 10 2 4 22 29 18 17 7 7 FNIR @ FPIR = 10−3 Table 34: Tabulation of median identification time results for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 14 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 125 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 126 — 6 4 58.92 21.97 20.15 25 5 4 83.71 22.27 21.10 18 7 4 15 16 0.0106 0.0018 0.0015 0.0085 0.0094 FNIR @ FPIR = 10−3 1 — 17 128.86 Total — 1 1.40 2 28 Stage Two 1 — 25 74.97 Stage One 21.10 1 2.42 2 27 Ten 4 22.27 28 1.79 One 20.15 5 81.93 1 26 Ten 5 21.97 25 1.86 One 1 6 56.37 1 25 Ten 2 19 126.56 One 1 2 28 0.0536 73.52 25 0.0536 2 25 0.0447 28 59.83 24 0.0333 2 13 72.22 21 0.0201 0.0050 19 7.67 20 12 55.79 1 18.47 9.82 16 68.20 3 44.54 20 7.78 26 9 — 1 19.46 8.63 1 — 3 42.73 1 — 30 11 — 1 — — 1 — 1 — 1 — 1 1 59.83 1 — — 14 72.22 1 — 20 7.67 1 1 55.79 1 18.47 9.82 18 68.20 3 44.54 2 22 7.78 26 9 1 1 19.46 8.63 2 3 0.0199 42.73 1 30 19 11 2 94.60 0.0013 95.02 1 0.0014 27 56.96 2 0.0051 84.51 12 30.34 13 0.0033 84.50 54.56 6 36.39 9 29 15 30.82 7 75.20 — — 7 33.02 23 — 1 — 9 67.98 1 — 1 — 19 — 1 — 1 — — 56.96 1 — 1 0.0097 1 12 30.34 1 — 17 94.60 54.56 6 36.39 1 15 95.02 2 17 30.82 7 75.20 8 27 1 7 33.02 23 1 84.51 2 9 67.98 1 0.0083 84.50 1 21 16 14 29 2 8 64.81 0.0783 66.39 28 0.0716 16 50.93 27 0.0034 31.68 10 52.10 11 0.0033 20.25 48.08 11 60.77 9 5 12 48.80 14 75.08 — — 13 54.42 22 — 1 — 14 69.30 1 — 1 — 21 — 1 — 1 — — 50.93 1 — 1 0.0017 1 10 52.10 1 — 5 64.80 48.08 11 60.77 1 21 57.57 2 12 48.80 15 75.08 26 13 1 14 54.41 22 27 31.68 2 16 69.30 26 19.64 1 23 19 4 2 27 0.0860 0.0014 29 0.2462 2 30 0.0358 73.83 77.04 23 0.0351 69.95 24 72.56 22 0.0017 18 69.71 20 NA 5 0.0019 74.40 22 70.77 — NA 8 74.22 — 23 82.61 — 42.25 25 1 — 28 72.48 8 67.33 2.49 — 1 NA 24 40.74 17 2.51 1 — — NA 10 65.07 28 77.04 1 34.90 — — 18 2.22 24 72.56 30 21.02 1 — 2.27 69.71 21 NA 29 — 1 27 2 24 70.77 — NA 1 — 71.41 1 25 48.12 — 42.25 1 67.31 2 13 52.49 8 67.32 17 1 15 40.74 18 72.31 2 10 65.07 72.05 1 20 26 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 35: Tabulation of median identification time results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 15 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 37.74 88.48 88.07 48.76 53.93 48.40 69.87 10 30 29 15 17 14 28 2 1 2 1 2 1 2 36.83 9 1 1 66.80 2 26 67.88 27 1 2 38.41 11 2 59.05 41.48 13 1 22 52.47 16 2 1 65.74 25 1 57.91 64.83 2 20 22.14 4 24 1 2 16.78 2 2 57.63 61.87 23 1 19 58.57 21 2 1 34.06 7 1 18.30 8.99 1 27.47 56.17 18 2 3 41.16 12 1 2 34.85 8 2 5 33.68 One 6 1 Sub. # NA — 22 73.98 50.40 NA — 12 92.90 95.29 36.93 39.07 74.41 64.96 59.52 58.09 48.54 58.46 75.26 41.85 45.62 67.66 73.89 77.97 26.40 20.00 62.74 60.32 35.00 9.76 109.60 69.81 36.23 35.85 Ten 26 27 7 8 23 18 15 13 11 14 24 9 10 19 21 25 3 2 17 16 4 1 28 20 6 5 Stage One 1 1 29 30 1 1 27 28 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 26 25 1 1 — — 20.09 33.16 — — 1.89 1.91 — — — — — — — — — — — — — — — — — — 1.04 0.60 — — One 1 1 — — 1 1 28 27 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 26 25 1 1 Stage Two — — — — NA NA — — 2.36 1.70 — — — — — — — — — — — — — — — — — — 1.16 0.58 Ten 26 14 27 28 29 30 10 11 24 20 18 17 3 5 25 9 12 15 23 22 4 2 21 19 7 1 16 13 8 6 69.87 48.40 74.56 82.88 88.07 88.48 39.37 39.76 66.80 59.05 57.91 57.63 20.18 27.47 67.88 38.41 41.48 52.47 65.74 64.83 22.14 16.78 61.87 58.57 34.06 8.99 57.36 41.67 34.85 33.68 One 22 11 — — 26 27 7 8 23 18 14 12 16 13 24 9 10 19 21 25 3 2 17 15 4 1 28 20 6 5 Total 73.98 50.40 NA NA 92.90 95.29 39.44 40.80 74.41 64.96 59.52 58.09 60.35 58.47 75.26 41.85 45.62 67.66 73.89 77.97 26.40 20.00 62.74 60.32 35.00 9.76 110.64 70.49 36.23 35.85 Ten 0.0149 0.0169 0.0028 0.0018 0.0137 0.0056 0.2514 0.2514 0.0649 0.0521 0.0291 0.0285 0.0014 0.0011 0.0071 0.0034 0.0136 0.0129 0.3067 0.2514 0.0041 0.0036 0.0020 0.0022 0.1017 0.2366 0.0378 0.0295 0.0052 0.0039 17 18 6 3 16 12 27 27 24 23 20 19 2 1 13 7 15 14 30 27 10 8 4 5 25 26 22 21 11 9 FNIR @ FPIR = 10−3 Table 36: Tabulation of median identification time results for Class C — Ten-Finger Plain-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are median durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 16 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 127 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 128 D FPVTE – F INGERPRINT M ATCHING Accuracy Time Tradeoff Detailed Tables with Mean Values In order to reduce the number of tables in the main body of the report (Section 8), this appendix contains tables that show the search times for each stage of identification, for both a single process and ten processes. The tables in this appendix report mean times. For readers interested in median times, please refer to Appendix C. Class A results are in Tables 37 through 42, Class B results are in Tables 43 through 46, and Class C results are in Tables 47 through 49. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.87 3.87 0.57 16.71 2.52 3.78 9.52 3.65 4.26 10.36 6 17 4 30 12 16 23 15 19 24 1 1 2 1 1 1 1 1 1.63 1.38 3.43 11.44 7.71 15.37 4.18 5.96 1.45 6.22 10 8 14 27 22 29 18 20 9 21 2 1 2 1 1 1 1 1 V 2 2 0.64 1 5 2 1 3.34 13 1 15.19 0.30 2 2 2 10.47 25 1 2.07 10.48 26 2 28 1.34 7 11 0.56 3 2 2 2 0.29 One 1 1 Sub. # U T S Q P O M L K J I H G F E D C Letter 20 7 21 19 29 22 27 13 9 10 5 28 11 15 1 26 25 8 4 24 18 17 23 14 12 30 3 16 6 2 Stage One 6.08 1.18 6.72 5.47 15.90 7.83 11.82 3.65 1.45 1.77 0.65 12.82 1.88 3.81 0.27 11.04 10.97 1.42 0.59 10.52 5.16 4.62 10.08 3.73 2.48 16.88 0.54 3.90 0.84 0.29 Ten 1 30 1 1 1 27 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 1 1 — 13.27 — — — 2.71 4.46 — — — — — — — — — — — — — — — — — — — — 3.65 — — One 1 30 1 1 1 28 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 27 1 1 Stage Two — — — 18.84 — — — 1.74 1.84 — — — — — — — — — — — — — — — — — — — — 1.69 Ten 19 26 18 16 28 23 29 13 8 9 5 27 10 12 2 24 25 7 3 22 17 14 21 15 11 30 4 20 6 1 6.22 14.72 5.96 4.18 15.37 10.42 15.90 3.43 1.38 1.63 0.64 15.19 2.07 3.34 0.30 10.47 10.48 1.34 0.56 10.36 4.26 3.65 9.52 3.78 2.52 16.71 0.57 7.52 0.87 0.29 One 19 30 20 17 28 21 27 12 8 9 5 26 10 14 1 25 24 7 4 23 16 15 22 13 11 29 3 18 6 2 Total 6.08 20.02 6.72 5.47 15.90 9.57 13.65 3.65 1.45 1.77 0.65 12.82 1.88 3.81 0.28 11.04 10.97 1.42 0.59 10.52 5.16 4.62 10.08 3.73 2.48 16.88 0.54 5.59 0.84 0.29 Ten 0.0226 0.0571 NA 0.0685 0.1218 0.0253 7 30 9 21 4 0.0222 3 0.1272 2 0.1308 23 22 0.0766 0.0818 15 13 0.2921 0.2995 28 0.0351 6 0.0625 8 29 0.0875 0.0883 17 16 0.0712 0.0786 10 0.0257 5 14 0.1576 26 0.1607 0.1089 19 27 0.1082 0.1111 18 0.0723 0.0745 12 20 0.0197 1 11 0.1337 0.1335 25 24 FNIR @ FPIR = 10−3 Table 37: Tabulation of mean identification time results for Class A — Left Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. < 20 seconds Participant FPVTE – F INGERPRINT M ATCHING 129 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 130 6 — 24.39 1 1 6 — — — 3.25 1 4 3 5 2 66.23 25.66 45.59 43.24 54.24 42.64 6 2 4 3 5 1 66.39 40.14 46.37 44.66 54.61 21.75 2 6 4 3 5 1 0.0252 0.1178 0.0650 0.0278 0.1086 0.0197 FNIR @ FPIR = 10−3 18.50 1 — 1 0.40 6 Total 1 54.61 1 — 5 — Stage Two 18.26 5 44.66 1 0.46 1 Stage One Sub. # 1 54.24 3 46.37 5 — Participant Letter 2 5 43.24 4 39.74 1 Ten D 2 3 45.59 2 66.39 One G 2 4 25.20 6 Ten I 2 2 66.23 One S 2 6 Ten U 2 One V Table 38: Tabulation of mean identification time results for Class A — Left Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. ≥ 20 seconds C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 0.76 3.72 0.45 15.89 2.52 3.71 10.12 3.58 4.13 10.19 6 17 3 30 12 16 23 15 19 24 1 1 2 1 1 1 1 1 1.48 1.38 3.49 11.10 7.43 15.08 3.82 5.64 1.30 5.95 10 9 14 27 22 29 18 20 8 21 2 1 2 1 1 1 1 1 V 2 2 0.59 1 5 2 1 3.26 13 1 11.31 0.29 2 2 2 10.30 25 1 1.75 10.42 26 2 28 1.25 7 11 0.53 4 2 2 2 0.25 One 1 1 Sub. # U T S Q P O M L K J I H G F E D C Letter 19 7 21 20 29 22 28 13 9 10 5 27 11 16 2 26 25 8 4 23 18 17 24 14 12 30 3 15 6 1 Stage One 5.88 1.06 6.83 6.12 15.12 7.66 11.54 3.48 1.37 1.67 0.62 10.80 1.71 3.74 0.27 10.73 10.65 1.33 0.55 10.06 5.13 4.61 10.32 3.52 2.37 16.40 0.51 3.72 0.77 0.27 Ten 1 30 1 1 1 27 28 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 29 1 1 — 10.72 — — — 2.71 3.34 — — — — — — — — — — — — — — — — — — — — 3.96 — — One 1 30 1 1 1 28 29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 27 1 1 Stage Two — — — 16.39 — — — 1.72 1.76 — — — — — — — — — — — — — — — — — — — — 1.63 Ten 19 27 18 16 29 22 28 13 8 9 5 26 10 12 2 24 25 7 4 23 17 14 21 15 11 30 3 20 6 1 5.95 12.02 5.64 3.82 15.08 10.15 14.44 3.49 1.38 1.48 0.59 11.31 1.75 3.26 0.29 10.30 10.42 1.25 0.53 10.19 4.13 3.58 10.12 3.71 2.52 15.89 0.45 7.68 0.76 0.25 One 18 30 20 19 28 21 27 12 8 9 5 26 10 14 2 25 24 7 4 22 16 15 23 13 11 29 3 17 6 1 Total 5.88 17.45 6.83 6.12 15.12 9.37 13.30 3.48 1.37 1.67 0.62 10.80 1.71 3.74 0.28 10.73 10.65 1.33 0.55 10.06 5.13 4.61 10.32 3.52 2.37 16.40 0.51 5.36 0.77 0.27 Ten 0.0996 0.0223 21 5 0.1929 28 0.0562 0.0442 7 9 0.0214 0.0218 2 0.1100 4 0.1133 25 22 0.0675 0.0776 17 13 0.2526 0.2615 29 0.0295 6 0.0505 8 30 0.0685 0.0682 14 15 0.0643 0.0708 12 0.0215 3 16 0.1230 26 0.1249 0.0910 19 27 0.0903 0.0933 18 0.0624 0.0630 11 20 0.0190 1 10 0.1124 0.1132 23 24 FNIR @ FPIR = 10−3 Table 39: Tabulation of mean identification time results for Class A — Right Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. < 20 seconds Participant FPVTE – F INGERPRINT M ATCHING 131 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 132 6 — 11.90 1 1 6 — — — 3.15 1 4 3 5 2 64.47 22.39 43.56 41.60 57.88 29.26 6 2 4 3 5 1 63.54 35.39 44.05 43.20 56.87 20.89 3 6 4 2 5 1 0.0222 0.1007 0.0503 0.0214 0.0909 0.0190 FNIR @ FPIR = 10−3 17.75 1 — 1 0.39 6 Total 1 56.87 1 — 5 — Stage Two 17.36 5 43.20 1 0.42 1 Stage One Sub. # 1 57.88 3 44.05 5 — Participant Letter 2 5 41.60 4 35.01 1 Ten D 2 3 43.56 2 63.54 One G 2 4 21.97 6 Ten I 2 2 64.47 One S 2 6 Ten U 2 One V Table 40: Tabulation of mean identification time results for Class A — Right Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. ≥ 20 seconds C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology < 20 seconds 1 1 1 1 1 1 I J K L O V 2.39 6.34 6.27 18.92 14.00 18.32 2.20 15.34 9.50 2 4 3 9 6 8 1 7 5 One 5 7 2 8 6 9 4 3 1 Stage One 9.59 16.58 5.03 18.82 15.16 20.61 7.48 6.55 2.66 Ten 1 1 1 1 1 1 1 1 1 — — — — — — — — — One 1 1 1 1 1 1 1 1 1 Stage Two — — — — — — — — — Ten 5 7 1 8 6 9 3 4 2 9.50 15.34 2.20 18.32 14.00 18.92 6.27 6.34 2.39 One 5 7 2 8 6 9 4 3 1 Total 9.59 16.58 5.03 18.82 15.16 20.61 7.48 6.55 2.66 Ten 0.0374 0.0515 0.0058 0.0143 0.0360 0.0146 0.0229 0.0034 9 2 3 6 4 5 1 0.0368 8 7 FNIR @ FPIR = 10−3 Table 41: Tabulation of mean identification time results for Class A — Left and Right Index — less than 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. 1 2 1 Sub. # G C Letter Participant FPVTE – F INGERPRINT M ATCHING C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 133 FPVTE – F INGERPRINT M ATCHING 134 16 234.52 73.01 27 2 21 5 23.15 190.00 37.95 10 11 4 4 0.0202 0.0207 0.0030 0.0030 FNIR @ FPIR = 10−3 23 22.27 Ten 18.74 2 Total 23 19.00 27 One 53.93 24 — Ten 24 70.74 1 Stage Two 19.21 26 1 One 1 170.99 — Ten 19.08 21 1 Stage One 1 163.78 1 One 1 21 23.15 Sub. # 2 3 0.0412 27 22 0.0311 22.27 15 0.0686 3 69.19 24 0.0684 27 13 244.70 23 0.0030 1 22 40.35 4 H M C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology Participant Letter D E 71.45 7 55.70 0.0386 15 227.27 10 21 22 37.65 24 57.53 — 8 53.77 0.0143 512.38 1 — 12 8 11 1 — 25 38.86 405.27 0.0286 59.30 — 1 — 6 14 0.0072 500.30 1 — 1 36.19 385.14 7 NA 13 1 — 1 7 32.99 26 NA — 69.19 1 — — 4 70.30 25 0.0214 — 13 244.70 1 — 32.68 14 31.32 12 1 23 40.35 1 1 6 23.54 3 174.39 — 71.45 8 55.70 — 3 32.59 8 19 — 16 227.27 10 — — 5 183.20 1 1 23 37.65 24 1 1 — 20 57.53 2 10 53.77 38.86 405.27 — 1 — 10 512.38 2 13 7 1 — 1 — 11 1 25 36.19 385.14 32.99 1 — 1 59.30 2 9 6 70.30 1 1 500.30 2 32.68 14 31.32 — 14 2 8 23.53 5 1 2 J I 2 4 32.59 1 F K 2 7 174.39 G L 1 9 22 0.0333 183.20 16 0.0027 22 1 0.0027 12 115.02 1 0.0281 2 16 247.48 13 0.0195 0.0370 23 180.13 9 20 114.37 20 19.75 50.30 17 213.08 1 62.27 21 163.65 26 12 — 19 20.11 45.52 1 81.88 1 63.41 26 85.04 26 14 — 27 — — 1 59.45 1 — 25 75.04 1 1 115.02 27 — — 17 165.60 1 — 20 95.09 1 1 114.37 16 19.75 50.30 18 153.63 2 62.27 1 20 88.60 26 12 2 17 20.11 45.52 1 2 NA 63.40 2 26 27 15 1 88.81 0.0366 432.22 19 0.0336 15 126.53 17 0.0358 26.96 17 55.33 18 0.0028 429.02 38.91 9 412.64 3 4 9 47.60 25 131.53 — — 11 252.04 18 — 1 26.86 24 133.45 1 — 25 0.72 18 — 1 18.77 22 — — 126.53 23 0.79 1 1 18 28.46 22 — 88.81 38.91 4 411.92 1 432.22 1 11 28.83 25 131.53 15 2 6 251.26 19 26.96 1 24 133.45 429.02 2 19 5 2 2 2 O P Q S T U V Table 42: Tabulation of mean identification time results for Class A — Left and Right Index — greater than or equal to 20-second searches. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. ≥ 20 seconds 47.51 78.95 79.01 33.07 42.62 36.28 48.01 16 27 28 10 14 11 18 2 1 2 1 2 1 2 47.81 17 1 1 78.80 2 26 80.54 29 1 2 26.88 7 2 38.17 51.59 21 1 12 63.06 25 2 1 52.05 22 1 90.51 49.13 20 2 30 62.98 24 1 2 19.65 6 2 32.39 46.18 15 1 9 31.92 8 2 1 41.57 13 1 5.51 3.71 1 12.52 55.36 23 2 5 48.97 19 1 2 6.62 4 2 3 3.71 One 2 1 Sub. # NA — 14 50.63 36.96 NA — 10 79.17 79.27 53.38 54.11 83.65 41.76 95.91 34.52 30.67 10.07 85.49 28.69 60.52 71.35 60.24 61.06 64.89 20.00 48.76 33.70 41.51 4.13 84.23 72.20 7.00 3.89 Ten 23 24 15 16 25 12 28 9 7 4 27 6 18 21 17 19 20 5 13 8 11 2 26 22 3 1 Stage One 1 1 29 30 1 1 26 25 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 — — 47.88 48.79 — — 3.54 3.52 — — — — — — — — — — — — — — — — — — 4.29 4.12 — — One 1 1 — — 1 1 26 25 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 3.35 3.24 — — — — — — — — — — — — — — — — — — 9.21 6.82 Ten 14 10 29 28 26 25 16 17 24 11 30 9 5 3 27 7 18 23 19 15 22 6 13 8 12 1 21 20 4 2 48.01 36.28 90.50 81.85 79.01 78.95 51.05 51.33 78.80 38.17 90.51 32.39 12.52 5.51 80.54 26.88 51.59 63.06 52.05 49.13 62.98 19.65 46.18 31.92 41.57 3.71 59.65 53.09 6.62 3.71 One 14 10 — — 23 24 15 16 25 12 28 9 7 4 26 6 18 21 17 19 20 5 13 8 11 2 27 22 3 1 Total 50.63 36.96 NA NA 79.17 79.27 56.74 57.34 83.65 41.76 95.91 34.52 30.67 10.08 85.49 28.69 60.52 71.35 60.24 61.06 64.89 20.00 48.76 33.70 41.51 4.13 93.44 79.02 7.00 3.89 Ten 0.0654 0.0647 0.0163 0.0142 0.0259 0.0187 0.1684 0.1681 0.0371 0.0325 0.0998 0.1008 0.0116 0.0094 0.0287 0.0236 0.0288 0.0276 0.1736 0.1634 0.0257 0.0254 0.0098 0.0099 0.1089 0.1133 0.0500 0.0461 0.0192 0.0190 22 21 6 5 13 7 29 28 18 17 23 24 4 1 15 10 16 14 30 27 12 11 2 3 25 26 20 19 9 8 FNIR @ FPIR = 10−3 Table 43: Tabulation of mean identification time results for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 10 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 135 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 136 — 4 1 53.78 6.54 3.70 24 3 1 79.03 6.74 3.74 13 2 6 23 24 0.0151 0.0052 0.0072 0.0392 0.0403 FNIR @ FPIR = 10−3 1 — 19 94.98 Total — 1 6.57 2 27 Stage Two 1 — 27 58.67 Stage One 3.74 1 3.71 2 20 Ten 1 6.74 27 9.49 One 3.70 3 72.46 1 28 Ten 1 6.54 22 4.69 One 1 4 50.07 1 28 Ten 2 19 85.48 One 1 2 27 0.1222 53.98 29 0.1220 2 28 0.0212 22 34.19 18 0.0198 2 8 49.37 16 0.0641 0.0083 14 12.33 25 7 32.28 5 44.17 4.14 8 46.58 13 40.45 14 11.80 18 11 — 5 41.12 4.68 1 — 13 40.27 1 — 17 12 — 1 — — 1 — 1 — 1 — 1 1 34.19 1 — — 8 49.37 1 — 14 12.33 1 1 32.28 5 44.17 4.14 8 46.58 13 40.45 2 16 11.80 18 11 1 5 41.12 4.68 2 14 0.0647 40.27 1 20 26 13 2 61.22 0.0058 60.09 5 0.0045 16 71.23 1 0.0156 50.97 21 60.26 14 0.0126 52.86 63.57 17 28.48 10 18 23 49.18 6 84.26 — — 16 25.86 26 — 1 — 7 77.67 1 — 1 — 27 — 1 — 1 — — 71.23 1 — 1 0.0167 1 21 60.26 1 — 15 61.22 63.57 17 28.48 1 4 60.09 2 25 49.18 6 84.26 3 16 1 18 25.86 26 1 50.97 2 7 77.67 1 0.0202 52.86 1 29 4 17 21 2 3 10.59 0.1259 31.95 30 0.1155 7 35.23 27 0.0142 5.49 9 96.58 12 0.0132 12.66 33.43 28 41.40 11 6 9 91.42 12 82.92 — — 30 38.08 25 — 1 — 11 77.37 1 — 1 — 26 — 1 — 1 — — 35.23 1 — 1 0.0057 1 9 96.58 1 — 3 10.59 33.43 28 41.40 1 20 31.95 2 10 91.42 12 82.92 22 7 1 30 38.08 25 25 5.49 2 12 77.37 26 12.66 1 28 20 6 2 24 0.0369 0.0057 21 0.0381 3 22 0.0266 68.15 76.78 19 0.0273 66.31 22 77.18 20 0.0106 19 74.29 23 NA 8 0.0110 62.95 25 73.96 — NA 9 61.29 — 24 89.71 — 37.19 21 1 — 29 88.88 10 50.37 3.29 — 1 NA 28 35.66 15 3.35 1 — — NA 10 48.47 26 76.78 1 56.35 — — 15 3.67 23 77.18 30 46.01 1 — 3.56 74.29 24 NA 29 — 1 25 2 27 73.96 — NA 1 — 64.86 1 26 33.35 — 37.19 1 62.96 2 9 42.87 10 50.37 19 1 15 35.66 15 59.28 2 11 48.47 57.74 1 17 23 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 44: Tabulation of mean identification time results for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 11 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 52.85 85.05 84.02 30.21 40.05 39.24 52.97 19 30 29 10 14 12 20 2 1 2 1 2 1 2 62.90 22 1 1 74.57 2 27 75.92 28 1 2 27.52 7 2 40.42 57.94 21 1 15 47.35 17 2 1 72.35 26 1 49.50 70.12 24 2 18 42.86 16 1 2 5.52 1 2 28.89 38.37 11 1 9 28.72 8 2 1 39.56 13 1 20.85 8.18 3 10.37 71.41 25 2 6 67.64 23 1 2 10.63 5 2 4 7.36 One 2 1 Sub. # NA — 16 55.45 41.22 NA — 11 87.27 87.49 61.10 71.42 78.24 43.46 52.63 30.58 63.37 22.54 80.17 29.91 73.10 53.29 84.76 85.94 43.96 5.43 40.60 30.42 40.38 7.40 124.70 120.11 11.08 7.86 Ten 25 26 17 19 21 12 14 8 18 5 22 6 20 15 23 24 13 1 10 7 9 2 28 27 4 3 Stage One 1 1 29 30 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 — — 50.99 56.68 — — 1.03 1.10 — — — — — — — — — — — — — — — — — — 2.80 1.72 — — One 1 1 — — 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 0.78 0.84 — — — — — — — — — — — — — — — — — — 5.36 3.14 Ten 17 11 30 29 27 28 18 20 25 13 16 9 6 4 26 7 19 15 23 22 14 1 10 8 12 3 24 21 5 2 52.97 39.24 91.04 86.88 84.02 85.05 53.88 63.99 74.57 40.42 49.50 28.89 20.85 10.37 75.92 27.52 57.94 47.35 72.35 70.12 42.86 5.52 38.37 28.72 39.56 8.18 74.21 69.35 10.63 7.36 One 16 11 — — 25 26 17 19 21 12 14 8 18 5 22 6 20 15 23 24 13 1 10 7 9 2 28 27 4 3 Total 55.45 41.22 NA NA 87.27 87.49 61.88 72.27 78.24 43.46 52.63 30.58 63.37 22.54 80.17 29.91 73.10 53.29 84.76 85.94 43.96 5.43 40.60 30.42 40.38 7.40 130.06 123.25 11.08 7.86 Ten NA NA 0.0031 0.0024 0.0063 0.0049 0.0910 0.0901 0.0106 0.0084 0.0349 0.0361 0.0022 0.0015 0.0068 0.0047 0.0054 0.0062 0.0904 0.0882 0.0057 0.0051 0.0021 0.0022 0.0160 0.0190 0.0139 0.0124 0.0036 0.0036 30 29 6 5 15 10 28 26 18 17 23 24 3 1 16 9 12 14 27 25 13 11 2 3 21 22 20 19 7 7 FNIR @ FPIR = 10−3 Table 45: Tabulation of mean identification time results for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 12 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 137 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 138 — 4 3 52.42 10.28 9.00 26 4 3 98.05 10.71 9.11 16 2 6 29 30 0.0043 0.0012 0.0020 NA NA FNIR @ FPIR = 10−3 1 — 17 105.02 Total — 1 1.80 2 28 Stage Two 1 — 27 54.37 Stage One 9.11 1 1.00 2 19 Ten 3 10.71 25 3.26 One 9.00 4 96.25 1 28 Ten 3 10.28 25 1.80 One 1 4 51.42 1 28 Ten 2 18 101.77 One 1 2 28 0.0591 52.57 27 0.0591 2 27 0.0062 20 40.86 18 0.0040 2 11 54.45 14 0.0203 0.0024 14 5.99 23 7 39.12 1 27.42 8.63 12 51.59 5 54.84 16 6.33 25 15 — 1 31.67 8.76 1 — 9 52.88 1 — 26 18 — 1 — — 1 — 1 — 1 — 1 1 40.86 1 — — 11 54.45 1 — 14 5.99 1 1 39.12 1 27.42 8.63 13 51.59 5 54.84 2 19 6.33 26 15 1 1 31.67 8.76 2 10 0.0204 52.88 1 27 24 21 2 97.35 0.0012 98.29 2 0.0009 27 29.56 1 0.0049 82.43 7 78.97 17 0.0033 85.74 24.57 20 28.07 11 27 6 60.01 6 69.51 — — 20 26.31 18 — 1 — 7 64.38 1 — 1 — 22 — 1 — 1 — — 29.56 1 — 1 0.0031 1 7 78.97 1 — 10 97.35 24.57 20 28.07 1 8 98.29 2 6 60.01 6 69.51 5 27 1 22 26.31 18 1 82.43 2 7 64.38 1 0.0033 85.74 1 24 8 11 28 2 5 34.20 0.0543 86.45 26 0.0515 22 43.38 25 0.0041 14.42 12 74.53 15 0.0035 28.56 41.37 19 39.70 13 8 13 70.45 10 68.54 — — 23 37.04 17 — 1 — 11 63.87 1 — 1 — 21 — 1 — 1 — — 43.38 1 — 1 0.0012 1 12 74.53 1 — 2 34.20 41.37 19 39.70 1 16 86.45 2 14 70.45 10 68.54 14 22 1 26 37.04 17 25 14.42 2 12 63.87 27 28.56 1 23 16 8 2 16 0.0108 0.0012 20 0.0136 2 21 0.0099 57.50 91.50 19 0.0141 82.97 23 93.61 22 0.0027 21 86.50 24 NA 9 0.0024 48.85 28 88.70 — NA 7 71.67 — 29 88.83 — 37.75 24 1 — 30 80.14 9 51.37 1.02 — 1 NA 25 35.51 13 1.05 1 — — NA 10 49.72 26 91.50 1 58.50 — — 15 1.47 23 93.61 30 37.63 1 — 1.46 86.50 24 NA 29 — 1 26 2 29 88.70 — NA 1 — 56.48 1 30 30.33 — 37.75 1 81.92 2 9 42.51 9 51.37 21 1 15 35.50 13 47.38 2 11 49.72 70.21 1 17 25 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 46: Tabulation of mean identification time results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 13 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 41.98 87.76 89.24 47.49 50.30 47.76 63.19 8 29 30 13 15 14 19 2 1 2 1 2 1 2 43.10 10 1 1 82.13 2 24 85.17 26 1 2 46.32 12 2 73.31 56.93 17 1 22 79.70 23 2 1 64.90 20 1 45.96 67.90 2 11 37.15 7 21 1 2 11.21 1 2 42.60 60.33 18 1 9 50.99 16 2 1 85.50 27 1 21.26 15.94 2 17.53 82.82 25 2 6 85.52 28 1 2 18.31 4 2 3 21.04 One 5 1 Sub. # NA — 16 66.29 49.30 NA — 11 92.85 97.78 45.57 47.65 91.94 80.40 45.71 42.17 67.26 39.66 94.45 50.85 64.06 84.04 72.52 77.93 48.42 14.65 60.33 50.83 88.32 15.93 166.16 173.53 19.18 22.92 Ten 24 26 7 9 23 20 8 6 17 5 25 13 15 21 18 19 10 1 14 12 22 2 27 28 3 4 Stage One 1 1 29 30 1 1 27 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 25 1 1 — — 17.68 27.65 — — 1.70 1.64 — — — — — — — — — — — — — — — — — — 2.09 1.55 — — One 1 1 — — 1 1 25 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 28 27 1 1 Stage Two — — — — NA NA — — 1.39 1.54 — — — — — — — — — — — — — — — — — — 2.98 2.52 Ten 17 13 20 22 30 29 9 10 24 21 11 8 6 3 26 12 15 23 18 19 7 1 16 14 27 2 25 28 4 5 63.19 47.76 67.98 75.14 89.24 87.76 43.67 44.75 82.13 73.31 45.96 42.60 21.26 17.53 85.17 46.32 56.93 79.70 64.90 67.91 37.15 11.21 60.34 50.99 85.50 15.94 84.92 87.07 18.31 21.04 One 16 11 — — 24 26 8 10 23 20 7 6 17 5 25 13 15 21 18 19 9 1 14 12 22 2 27 28 3 4 Total 66.29 49.30 NA NA 92.85 97.78 46.96 49.18 91.94 80.40 45.71 42.17 67.27 39.67 94.45 50.85 64.06 84.04 72.52 77.93 48.42 14.65 60.33 50.83 88.32 15.93 169.13 176.06 19.18 22.92 Ten NA 0.0711 0.0015 0.0011 0.0088 0.0048 0.0734 0.0734 0.0368 0.0276 0.0276 0.0275 0.0013 0.0010 0.0047 0.0027 0.0102 0.0095 0.0934 0.0826 0.0025 0.0027 0.0011 0.0013 0.0311 0.1680 0.0163 0.0155 0.0024 0.0024 30 24 6 2 14 13 25 25 23 20 20 19 4 1 12 10 16 15 28 27 9 10 2 4 22 29 18 17 7 7 FNIR @ FPIR = 10−3 Table 47: Tabulation of mean identification time results for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 14 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 139 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 140 — 5 4 65.97 25.91 25.25 27 4 3 104.36 27.89 27.50 18 7 4 15 16 0.0106 0.0018 0.0015 0.0085 0.0094 FNIR @ FPIR = 10−3 1 — 17 163.56 Total — 1 1.91 1 28 Stage Two 1 — 25 86.39 Stage One 27.50 1 4.17 1 30 Ten 3 27.89 28 3.02 One 25.25 4 102.45 1 28 Ten 4 25.91 27 2.81 One 1 5 61.79 1 27 Ten 2 18 160.54 One 1 1 28 0.0536 83.58 25 0.0536 1 25 0.0447 28 58.39 24 0.0333 2 11 71.58 21 0.0201 0.0050 18 12.99 20 12 55.63 2 31.33 59.56 13 68.61 5 10.54 18 11.67 25 13 — 2 30.61 9.52 1 — 6 57.02 1 — 28 14 — 1 — — 1 — 1 — 1 — 1 1 58.39 1 — — 11 71.58 1 — 18 12.99 1 1 55.63 2 31.33 59.56 15 68.61 5 10.54 2 20 11.67 25 13 1 2 30.61 9.52 2 6 0.0199 57.01 1 29 19 16 2 94.22 0.0013 94.72 1 0.0014 26 75.61 2 0.0051 84.14 19 41.07 13 0.0033 84.26 79.04 7 40.14 9 29 24 40.53 6 81.37 — — 9 36.09 22 — 1 — 8 72.62 1 — 1 — 20 — 1 — 1 — — 75.61 1 — 1 0.0097 1 19 41.07 1 — 17 94.22 79.04 7 40.14 1 14 94.72 2 25 40.53 6 81.37 7 26 1 9 36.09 23 1 84.14 2 8 72.62 1 0.0083 84.26 1 22 14 14 30 2 7 64.19 0.0783 58.45 28 0.0716 12 49.47 27 0.0034 31.42 9 50.67 11 0.0033 19.52 47.38 10 66.17 9 3 11 48.71 15 81.31 — — 12 59.01 21 — 1 — 15 72.55 1 — 1 — 19 — 1 — 1 — — 49.47 1 — 1 0.0017 1 9 50.67 1 — 5 64.18 47.38 10 66.17 1 24 58.45 2 11 48.71 15 81.31 26 12 1 13 59.01 22 26 31.42 2 17 72.55 25 19.51 1 21 24 3 2 26 0.0860 0.0014 29 0.2462 2 30 0.0358 85.39 80.86 23 0.0351 83.58 20 71.01 22 0.0017 23 74.01 17 NA 5 0.0019 83.25 22 72.95 — NA 8 83.35 — 21 82.76 — 42.49 27 1 — 25 74.39 8 67.92 2.53 — 1 NA 23 40.94 16 2.55 1 — — NA 10 65.47 27 80.86 1 34.24 — — 16 2.23 20 71.01 30 21.36 1 — 2.30 74.01 17 NA 29 — 1 26 2 24 72.95 — NA 1 — 82.87 1 23 48.52 — 42.49 1 81.03 2 12 53.02 8 67.92 21 1 14 40.94 16 81.02 2 10 65.47 81.05 1 19 27 2 1 1 1 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 48: Tabulation of mean identification time results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 15 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 47.87 89.37 89.34 48.44 53.32 48.91 71.28 10 30 29 11 13 12 24 2 1 2 1 2 1 2 47.17 9 1 1 75.98 2 26 79.37 28 1 2 42.13 8 2 65.71 66.90 23 1 22 76.98 27 2 1 65.69 21 1 57.90 65.03 2 16 34.47 6 20 1 2 25.05 3 2 57.06 61.82 19 1 15 58.39 17 2 1 59.20 18 1 18.62 10.81 1 27.43 74.38 25 2 2 55.60 14 1 2 33.93 5 2 4 36.55 One 7 1 Sub. # NA — 19 74.86 50.94 NA — 8 90.28 96.29 51.32 52.18 86.80 74.07 58.12 56.59 50.64 58.35 89.84 46.98 76.96 82.78 73.62 78.10 45.09 32.02 61.34 58.89 61.17 11.40 155.17 98.54 35.83 40.68 Ten 25 26 9 10 23 18 12 11 7 13 24 6 20 22 17 21 5 2 16 14 15 1 28 27 3 4 Stage One 1 1 29 30 1 1 27 28 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 26 25 1 1 — — 20.14 34.07 — — 1.85 1.91 — — — — — — — — — — — — — — — — — — 1.62 0.77 — — One 1 1 — — 1 1 28 26 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 27 25 1 1 Stage Two — — — — NA NA — — 2.36 1.72 — — — — — — — — — — — — — — — — — — 1.90 0.71 Ten 22 9 23 28 29 30 11 10 24 20 14 13 2 4 27 8 21 26 19 18 6 3 17 15 16 1 25 12 5 7 71.28 48.91 73.46 82.50 89.34 89.37 49.72 49.07 75.98 65.71 57.90 57.06 18.62 27.43 79.37 42.13 66.90 76.98 65.69 65.03 34.47 25.05 61.82 58.39 59.20 10.81 76.00 56.38 33.93 36.55 One 19 8 — — 25 26 9 10 23 18 12 11 7 13 24 6 20 22 17 21 5 2 16 14 15 1 28 27 3 4 Total 74.86 50.94 NA NA 90.28 96.29 53.68 53.90 86.80 74.07 58.12 56.59 50.65 58.36 89.84 46.98 76.96 82.78 73.62 78.10 45.09 32.02 61.34 58.89 61.17 11.40 157.06 99.25 35.83 40.68 Ten 0.0149 0.0169 0.0028 0.0018 0.0137 0.0056 0.2514 0.2514 0.0649 0.0521 0.0291 0.0285 0.0014 0.0011 0.0071 0.0034 0.0136 0.0129 0.3067 0.2514 0.0041 0.0036 0.0020 0.0022 0.1017 0.2366 0.0378 0.0295 0.0052 0.0039 17 18 6 3 16 12 27 27 24 23 20 19 2 1 13 7 15 14 30 27 10 8 4 5 25 26 22 21 11 9 FNIR @ FPIR = 10−3 Table 49: Tabulation of mean identification time results for Class C — Ten-Finger Plain-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Stage One and Stage Two refer to the stages of identification defined in Section 5, with Total indicating the combined search time. One and Ten refer to the number of concurrent identification processes run on a compute node. All values are mean durations are reported in seconds, but were originally recorded to microsecond precision. A — indicates that the operation completed faster than could be reliably measured. NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 16 are reprinted to the right of this table. V U S Q O M L J I H G F E D C Letter Participant FPVTE – F INGERPRINT M ATCHING 141 C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 142 E FPVTE – F INGERPRINT M ATCHING Progression for Last Two Submissions This appendix provides additional information in reference to comments from Section 8 that looked at accuracy versus search time. The tables in this appendix show search times for the last two of the three submission periods during the evaluation. Generally, participants attempt to improve accuracy with potential tradeoffs in speed and template size. The tables in this section show the search times and FNIR @ FPIR = 10−3 for the these two submissions. Class A results are in Tables 50 through 52, Class B results are in Tables 53 through 56, and Class C results are in Tables 57 through 59. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D Third FNIR 1 1 0.12 16 0.1077 2 6 0.75 15 1 9 1.79 2 Time FNIR 1 0.29 24 0.1335 0.0984 6 0.87 25 0.1337 0.0376 20 7.52 1 0.0197 3 0.42 10 0.0794 4 0.57 12 0.0745 2 22 15.40 8 0.0757 30 16.71 11 0.0723 1 11 2.50 18 0.1122 11 2.52 20 0.1111 2 15 3.64 17 0.1093 15 3.78 18 0.1082 1 16 5.30 23 0.1221 21 9.52 19 0.1089 1 5 0.71 20 0.1168 14 3.65 26 0.1576 2 7 0.94 19 0.1160 17 4.26 27 0.1607 1 18 6.19 1 0.0306 22 10.36 5 0.0257 1 2 0.35 13 0.0900 3 0.56 14 0.0786 2 4 0.60 9 0.0773 7 1.34 10 0.0712 1 — — — — 25 10.48 17 0.0883 2 — — — — 24 10.47 16 0.0875 1 8 1.76 14 0.0913 2 0.30 8 0.0625 2 14 3.04 12 0.0881 12 3.34 6 0.0351 1 — — — — 10 2.07 29 0.2995 2 — — — — 27 15.19 28 0.2921 1 10 1.96 7 0.0751 5 0.64 15 0.0818 2 13 2.95 6 0.0735 9 1.63 13 0.0766 1 — — — — 8 1.38 23 0.1308 2 12 2.83 24 0.1343 13 3.43 22 0.1272 1 20 6.97 4 0.0507 29 15.90 2 0.0222 2 19 6.42 5 0.0511 23 10.42 3 0.0226 1 17 5.99 11 0.0811 28 15.37 7 0.0571 1 23 16.49 21 0.1181 16 4.18 30 NA 2 — — — — 18 5.96 9 0.0685 U 1 24 18.62 22 0.1209 26 14.72 21 0.1218 V 1 21 10.34 3 0.0402 19 6.22 4 0.0253 D 2 1 3.99 1 0.0269 2 42.64 1 0.0197 G 2 4 27.35 5 0.1210 5 54.24 5 0.1086 I 2 3 9.04 2 0.0274 3 43.24 3 0.0278 F G H I < 20 seconds Second Time Sub. # 1 E J K L M O P Q S T ≥ 20 seconds 143 S 2 2 5.94 4 0.0811 4 45.59 4 0.0650 U 2 — — — — 1 25.66 6 0.1178 V 2 5 61.50 3 0.0395 6 66.23 2 0.0252 Table 50: Tabulation of the progression of identification timing and accuracy for Class A — Left Index. Submissions were split into two groups. The first group includes submissions that, in Third, performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 100 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 144 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H < 20 seconds I J K L M O Third FNIR 1 1 0.11 20 0.1022 2 6 0.67 15 1 9 1.77 5 Time FNIR 1 0.25 24 0.1132 0.0899 6 0.76 23 0.1124 0.0337 20 7.68 1 0.0190 1 2 0.37 12 0.0733 3 0.45 11 0.0630 2 23 14.66 10 0.0709 30 15.89 10 0.0624 1 11 2.26 18 0.0945 11 2.52 20 0.0933 2 15 3.30 17 0.0928 15 3.71 18 0.0903 1 16 5.39 19 0.1018 21 10.12 19 0.0910 1 5 0.65 22 0.1035 14 3.58 26 0.1230 2 7 0.88 21 0.1025 17 4.13 27 0.1249 1 17 5.91 1 0.0232 23 10.19 3 0.0215 1 3 0.37 14 0.0800 4 0.53 16 0.0708 2 4 0.55 9 0.0700 7 1.25 12 0.0643 1 — — — — 25 10.42 14 0.0682 2 — — — — 24 10.30 15 0.0685 1 8 1.54 13 0.0739 2 0.29 8 0.0505 2 14 2.94 11 0.0721 12 3.26 6 0.0295 1 — — — — 10 1.75 30 0.2615 2 — — — — 26 11.31 29 0.2526 1 10 1.89 7 0.0672 5 0.59 17 0.0776 2 13 2.79 8 0.0673 9 1.48 13 0.0675 1 — — — — 8 1.38 25 0.1133 2 12 2.62 24 0.1120 13 3.49 22 0.1100 1 22 12.96 3 0.0248 28 14.44 4 0.0218 2 19 9.23 2 0.0242 22 10.15 2 0.0214 1 18 6.18 6 0.0640 29 15.08 7 0.0442 1 21 12.30 16 0.0924 16 3.82 28 0.1929 2 — — — — 18 5.64 9 0.0562 1 24 15.97 23 0.1048 27 12.02 21 0.0996 V 1 20 9.96 4 0.0331 19 5.95 5 0.0223 D 2 2 6.25 2 0.0248 2 29.26 1 0.0190 G 2 4 26.76 5 0.1007 5 57.88 5 0.0909 I 2 3 9.31 1 0.0217 3 41.60 2 0.0214 S 2 1 6.20 4 0.0641 4 43.56 4 0.0503 U 2 — — — — 1 22.39 6 0.1007 V 2 5 60.50 3 0.0330 6 64.47 3 0.0222 P Q S T U ≥ 20 seconds Second Time Sub. # Table 51: Tabulation of the progression of identification timing and accuracy for Class A — Right Index. Submissions were split into two groups. The first group includes submissions that, in Third, performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 100 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter < 20 seconds C Second Time Sub. # Third FNIR Time FNIR 1 1 1.48 6 0.0313 2 2.39 7 0.0368 2 2 6.30 7 0.0354 4 6.34 8 0.0374 G 1 3 7.38 8 0.1426 3 6.27 9 0.0515 I 1 7 19.73 2 0.0073 9 18.92 2 0.0058 J 1 5 9.20 4 0.0161 6 14.00 3 0.0143 K 1 — — — — 8 18.32 6 0.0360 L 1 6 17.81 5 0.0306 1 2.20 4 0.0146 O 1 8 57.36 3 0.0132 7 15.34 5 0.0229 V 1 4 8.56 1 0.0056 5 9.50 1 0.0034 1 2 14.06 5 0.0046 16 73.01 4 0.0030 2 9 46.57 1 0.0033 23 234.52 4 0.0030 1 4 17.36 10 0.0216 2 22.27 11 0.0207 2 22 463.95 9 0.0208 27 500.30 10 0.0202 1 12 56.49 15 0.0392 13 59.30 21 0.0386 2 13 67.36 16 0.0407 15 71.45 22 0.0412 2 19 166.37 21 0.0558 22 227.27 15 0.0311 1 5 18.42 20 0.0538 8 37.65 24 0.0686 2 6 25.38 19 0.0537 12 53.77 23 0.0684 I 2 18 162.34 4 0.0038 25 385.14 4 0.0030 J 2 3 16.23 8 0.0137 7 36.19 8 0.0143 K 2 — — — — 6 32.68 14 0.0286 L 2 7 34.31 13 0.0365 3 23.54 7 0.0072 1 — — — — 5 32.59 26 NA 2 — — — — 20 183.20 25 NA 2 15 85.37 7 0.0134 10 45.52 12 0.0214 1 — — — — 14 63.41 20 0.0370 2 16 89.78 17 0.0408 17 114.37 16 0.0333 1 14 85.29 3 0.0035 21 213.08 1 0.0027 2 10 48.40 2 0.0034 19 163.65 1 0.0027 1 1 13.37 14 0.0379 1 20.11 13 0.0281 2 11 54.11 11 0.0316 26 429.02 9 0.0195 1 17 138.43 12 0.0318 4 26.96 27 NA 2 — — — — 9 38.91 19 0.0366 D E F G H ≥ 20 seconds 145 M O P Q S T U V 1 8 38.87 18 0.0532 11 47.60 17 0.0336 2 21 288.82 22 0.2620 24 252.04 18 0.0358 2 20 197.77 6 0.0048 18 133.45 3 0.0028 Table 52: Tabulation of the progression of identification timing and accuracy for Class A — Left and Right Index. Submissions were split into two groups. The first group includes submissions that, in Third, performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 1 600 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 146 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Second Time Sub. # 1 1 Third FNIR 3.20 14 0.0667 Time 2 FNIR 3.71 22 0.0654 2 3 6.18 15 0.0669 4 6.62 21 0.0647 1 14 39.90 5 0.0178 20 53.09 6 0.0163 2 18 55.83 7 0.0242 21 59.65 5 0.0142 1 2 3.37 12 0.0397 1 3.71 13 0.0259 2 7 19.68 8 0.0277 12 41.57 7 0.0187 1 — — — — 8 31.92 29 0.1684 2 — — — — 13 46.18 28 0.1681 1 12 33.42 17 0.0704 6 19.65 18 0.0371 2 19 61.44 16 0.0677 22 62.98 17 0.0325 1 13 39.16 18 0.0798 15 49.13 23 0.0998 2 17 46.21 20 0.0813 19 52.05 24 0.1008 1 21 77.87 4 0.0161 23 63.06 4 0.0116 2 20 67.21 3 0.0127 18 51.59 1 0.0094 1 5 11.15 13 0.0491 7 26.88 15 0.0287 2 11 31.32 9 0.0340 27 80.54 10 0.0236 1 4 10.66 19 0.0809 3 5.51 16 0.0288 2 8 23.68 20 0.0813 5 12.52 14 0.0276 1 — — — — 9 32.39 30 0.1736 2 22 89.64 22 0.1634 30 90.51 27 0.1634 1 6 19.29 11 0.0366 11 38.17 12 0.0257 2 9 30.71 10 0.0349 24 78.80 11 0.0254 1 16 45.67 1 0.0104 17 51.33 2 0.0098 2 10 31.26 2 0.0116 16 51.05 3 0.0099 1 — — — — 25 78.95 25 0.1089 2 — — — — 26 79.01 26 0.1133 1 — — — — 28 81.85 20 0.0500 2 — — — — 29 90.50 19 0.0461 1 15 42.01 6 0.0239 10 36.28 9 0.0192 2 — — — — 14 48.01 8 0.0190 Table 53: Tabulation of the progression of identification timing and accuracy for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 3 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Second Time Sub. # 1 147 1 Third FNIR 3.16 16 0.0391 Time 1 FNIR 3.70 24 0.0403 2 3 6.10 14 0.0386 4 6.54 23 0.0392 1 13 37.74 5 0.0082 19 53.78 6 0.0072 2 19 56.55 6 0.0100 20 58.67 2 0.0052 1 2 3.71 11 0.0220 2 4.68 13 0.0151 2 9 25.48 7 0.0134 12 40.27 7 0.0083 1 — — — — 8 32.28 29 0.1222 2 — — — — 14 46.58 28 0.1220 1 6 19.93 17 0.0423 5 11.80 18 0.0212 2 15 41.42 15 0.0388 13 41.12 16 0.0198 1 14 39.40 20 0.0554 17 50.97 25 0.0641 2 17 46.02 21 0.0563 18 52.86 26 0.0647 1 21 79.29 4 0.0074 23 63.57 5 0.0058 2 20 77.42 1 0.0057 16 49.18 1 0.0045 1 5 11.62 13 0.0313 7 25.86 14 0.0156 2 11 32.12 9 0.0210 27 77.67 10 0.0126 1 4 11.09 18 0.0526 3 5.49 15 0.0167 2 8 23.42 19 0.0536 6 12.66 17 0.0202 1 — — — — 9 33.43 30 0.1259 2 22 90.28 22 0.1157 30 91.42 27 0.1155 1 7 20.12 12 0.0231 11 38.08 12 0.0142 2 10 32.04 10 0.0212 26 77.37 11 0.0132 1 18 53.98 2 0.0064 22 62.95 3 0.0057 2 12 33.57 3 0.0069 21 61.29 3 0.0057 1 — — — — 25 74.29 21 0.0369 2 — — — — 24 73.96 22 0.0381 1 — — — — 29 89.71 19 0.0266 2 — — — — 28 88.88 20 0.0273 1 16 42.11 8 0.0148 10 35.66 8 0.0106 2 — — — — 15 48.47 9 0.0110 Table 54: Tabulation of the progression of identification timing and accuracy for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 3 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 148 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Second Time Sub. # 1 1 Third FNIR 3.44 21 0.0365 Time FNIR 2 7.36 30 NA 2 3 4.33 20 0.0356 5 10.63 29 NA 1 13 38.51 5 0.0040 21 69.35 6 0.0031 2 17 48.86 6 0.0045 24 74.21 5 0.0024 1 4 6.39 11 0.0098 3 8.18 15 0.0063 2 18 52.30 8 0.0074 12 39.56 10 0.0049 1 — — — — 8 28.72 28 0.0910 2 — — — — 10 38.37 26 0.0901 1 2 4.27 13 0.0132 1 5.52 18 0.0106 2 7 22.15 16 0.0235 14 42.86 17 0.0084 1 19 52.34 18 0.0325 22 70.12 23 0.0349 2 20 61.14 19 0.0340 23 72.35 24 0.0361 1 22 79.56 4 0.0031 15 47.35 3 0.0022 2 21 69.84 3 0.0027 19 57.94 1 0.0015 1 5 11.06 15 0.0228 7 27.52 16 0.0068 2 12 32.23 10 0.0090 26 75.92 9 0.0047 1 8 25.03 17 0.0312 4 10.37 12 0.0054 2 10 26.13 14 0.0216 6 20.85 14 0.0062 1 — — — — 9 28.89 27 0.0904 2 16 48.74 22 0.0882 16 49.50 25 0.0882 1 6 19.15 12 0.0106 13 40.42 13 0.0057 2 11 31.55 9 0.0088 25 74.57 11 0.0051 1 15 47.54 1 0.0019 20 63.99 2 0.0021 2 9 25.99 2 0.0020 18 53.88 3 0.0022 1 — — — — 28 85.05 21 0.0160 2 — — — — 27 84.02 22 0.0190 1 — — — — 29 86.88 20 0.0139 2 — — — — 30 91.04 19 0.0124 1 14 46.90 7 0.0051 11 39.24 7 0.0036 2 — — — — 17 52.97 7 0.0036 Table 55: Tabulation of the progression of identification timing and accuracy for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 3 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V 149 Second Time Sub. # 1 2 2 1 Third FNIR 4.67 20 0.0276 3 5.01 19 9 26.83 6 2 13 33.98 1 4 7.71 2 20 70.09 1 — — 2 — — 1 1 2.50 2 6 1 18 Time FNIR 3 9.00 30 0.0268 4 10.28 29 NA 0.0028 17 52.42 6 0.0020 5 0.0027 19 54.37 2 0.0012 11 0.0071 2 8.76 16 0.0043 8 0.0043 18 52.88 7 0.0024 — — 12 39.12 27 0.0591 — — 16 51.59 27 0.0591 12 0.0101 1 6.33 18 0.0062 12.62 14 0.0157 9 31.67 14 0.0040 56.70 16 0.0220 26 82.43 23 0.0203 2 19 67.68 17 0.0249 27 85.74 24 0.0204 1 22 75.61 4 0.0021 6 24.57 2 0.0012 2 17 50.74 2 0.0012 20 60.01 1 0.0009 1 5 8.13 22 0.0614 7 26.31 17 0.0049 2 11 31.60 9 0.0054 22 64.38 11 0.0033 1 14 36.59 18 0.0250 5 14.42 10 0.0031 2 12 33.50 13 0.0147 8 28.56 11 0.0033 NA 1 — — — — 13 41.37 26 0.0543 2 21 70.37 21 0.0515 23 70.45 25 0.0515 1 7 16.34 15 0.0173 11 37.04 15 0.0041 2 10 30.74 10 0.0056 21 63.87 13 0.0035 1 16 48.59 1 0.0010 14 48.85 2 0.0012 2 8 26.38 2 0.0012 24 71.67 2 0.0012 1 — — — — 28 86.50 20 0.0108 2 — — — — 29 88.70 21 0.0136 1 — — — — 30 88.83 19 0.0099 2 — — — — 25 80.14 22 0.0141 1 15 42.75 7 0.0041 10 35.51 9 0.0027 2 — — — — 15 49.72 7 0.0024 Table 56: Tabulation of the progression of identification timing and accuracy for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 3 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 150 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Second Time Sub. # Third FNIR 3 10.27 16 0.0574 2 2 10.04 17 1 10 44.21 4 2 20 74.46 1 6 27.61 2 21 79.95 1 12 50.99 2 17 60.34 1 1 8.43 2 5 1 18 Time FNIR 5 21.04 30 NA 0.0584 4 18.31 24 0.0711 0.0024 28 87.07 6 0.0015 3 0.0018 25 84.92 2 0.0011 11 0.0143 2 15.94 14 0.0088 9 0.0083 27 85.50 13 0.0048 18 0.0734 14 50.99 25 0.0734 18 0.0734 16 60.34 25 0.0734 21 0.1923 1 11.21 23 0.0368 21.36 20 0.0780 7 37.15 20 0.0276 68.30 15 0.0377 19 67.91 20 0.0276 2 15 59.28 14 0.0363 18 64.90 19 0.0275 1 14 56.55 5 0.0033 23 79.70 4 0.0013 2 9 40.52 7 0.0050 15 56.93 1 0.0010 1 7 35.33 10 0.0104 12 46.32 12 0.0047 2 19 69.82 7 0.0050 26 85.17 10 0.0027 1 13 54.67 12 0.0159 3 17.53 16 0.0102 2 11 47.51 13 0.0221 6 21.26 15 0.0095 1 — — — — 8 42.60 28 0.0934 2 — — — — 11 45.96 27 0.0826 1 — — — — 21 73.31 9 0.0025 2 — — — — 24 82.13 10 0.0027 1 8 38.03 1 0.0014 10 44.75 2 0.0011 2 4 20.02 2 0.0017 9 43.67 4 0.0013 1 — — — — 29 87.76 22 0.0311 2 — — — — 30 89.24 29 0.1680 1 — — — — 22 75.14 18 0.0163 2 — — — — 20 67.98 17 0.0155 1 16 59.38 6 0.0034 13 47.76 7 0.0024 2 — — — — 17 63.19 7 0.0024 1 Table 57: Tabulation of the progression of identification timing and accuracy for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 5 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V 151 Second Time Sub. # Third FNIR 2 19.26 21 0.0826 2 7 28.17 20 1 11 46.42 6 2 16 70.86 1 1 12.49 2 8 29.59 1 13 55.63 2 15 68.61 1 3 20.17 2 5 1 19 2 1 Time FNIR 4 25.25 16 0.0094 0.0822 5 25.91 15 0.0085 0.0028 17 65.97 4 0.0015 5 0.0023 30 86.39 7 0.0018 17 0.0285 1 9.52 18 0.0106 9 0.0121 14 57.02 12 0.0050 18 0.0536 13 55.63 25 0.0536 18 0.0536 18 68.61 25 0.0536 15 0.0239 2 11.67 24 0.0447 23.05 13 0.0195 6 30.61 21 0.0333 77.49 15 0.0239 28 84.14 20 0.0201 17 71.01 14 0.0237 29 84.26 19 0.0199 10 40.06 1 0.0015 24 79.04 1 0.0013 2 4 20.92 7 0.0032 9 40.53 2 0.0014 1 6 26.55 12 0.0168 8 36.09 13 0.0051 2 12 55.60 8 0.0045 20 72.62 9 0.0033 1 18 75.36 10 0.0123 7 31.42 17 0.0097 2 20 80.46 11 0.0160 3 19.52 14 0.0083 1 — — — — 11 47.38 28 0.0783 2 — — — — 12 48.71 27 0.0716 1 — — — — 15 59.01 11 0.0034 2 — — — — 19 72.55 9 0.0033 1 14 67.74 1 0.0015 26 83.25 5 0.0017 2 9 37.87 3 0.0019 27 83.35 2 0.0014 1 — — — — 22 74.01 29 0.0860 2 — — — — 21 72.95 30 0.2462 1 — — — — 25 82.76 23 0.0358 2 — — — — 23 74.39 22 0.0351 1 21 91.48 4 0.0022 10 40.94 5 0.0017 2 — — — — 16 65.47 8 0.0019 1 Table 58: Tabulation of the progression of identification timing and accuracy for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 5 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 152 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Second Time Sub. # Third FNIR 1 3 18.23 16 0.0979 2 4 19.43 17 1 8 39.02 6 2 14 58.60 1 5 20.29 2 10 55.20 1 13 58.39 2 16 61.82 1 1 11.46 2 2 1 20 2 1 Time FNIR 7 36.55 17 0.0149 0.1124 5 33.93 18 0.0169 0.0048 12 56.38 6 0.0028 2 0.0034 25 76.00 3 0.0018 13 0.0383 1 10.81 16 0.0137 9 0.0164 16 59.20 12 0.0056 20 0.2514 15 58.39 27 0.2514 20 0.2514 17 61.82 27 0.2514 19 0.2393 3 25.05 24 0.0649 11.62 18 0.1455 6 34.47 23 0.0521 70.66 15 0.0429 18 65.03 20 0.0291 18 62.51 14 0.0411 19 65.69 19 0.0285 11 56.80 5 0.0039 26 76.98 2 0.0014 2 12 58.34 4 0.0037 21 66.90 1 0.0011 1 7 30.56 10 0.0174 8 42.13 13 0.0071 2 17 62.05 8 0.0053 27 79.37 7 0.0034 1 15 60.76 11 0.0251 4 27.43 15 0.0136 2 21 72.40 12 0.0358 2 18.62 14 0.0129 1 — — — — 13 57.06 30 0.3067 2 — — — — 14 57.90 27 0.2514 1 — — — — 20 65.71 10 0.0041 2 — — — — 24 75.98 8 0.0036 1 9 42.45 1 0.0028 10 49.07 4 0.0020 2 6 22.94 2 0.0034 11 49.72 5 0.0022 1 — — — — 30 89.37 25 0.1017 2 — — — — 29 89.34 26 0.2366 1 — — — — 28 82.50 22 0.0378 2 — — — — 23 73.46 21 0.0295 1 19 66.49 7 0.0051 9 48.91 11 0.0052 2 — — — — 22 71.28 9 0.0039 Table 59: Tabulation of the progression of identification timing and accuracy for Class C — Ten-Finger Plain-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The Time column shows the time used to perform a search over an enrollment set of 5 000 000 in seconds, but was originally recorded to microsecond precision. The FNIR column shows FNIR for each submission at FPIR = 10−3 . — indicates that there was not a validated submission during that submission period. The number to the left of a value provides the value’s column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING F Enrollment Size C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 153 154 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V RAM Actual Sub. # On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 1 7 0.18 10 0.16 8 0.17 8 0.16 30 0.1335 2 8 0.18 10 0.16 8 0.17 8 0.16 31 0.1337 1 21 0.61 18 0.33 29 1.25 18 0.34 1 0.0197 2 21 0.61 18 0.33 30 1.40 18 0.34 1 0.0197 1 30 1.12 27 0.50 23 0.50 27 0.50 16 0.0745 2 29 1.11 27 0.50 23 0.50 27 0.50 15 0.0723 1 6 0.18 7 0.11 10 0.17 10 0.17 25 0.1111 2 5 0.18 7 0.11 10 0.17 10 0.17 22 0.1082 1 20 0.57 29 0.50 5 0.15 5 0.15 24 0.1089 0.1086 2 19 0.57 29 0.50 5 0.15 5 0.15 23 1 11 0.33 14 0.29 12 0.29 14 0.29 32 0.1576 2 11 0.33 14 0.29 12 0.29 14 0.29 33 0.1607 1 23 0.70 35 0.68 31 1.54 35 0.68 7 0.0257 2 28 1.04 36 1.03 32 1.55 36 1.03 8 0.0278 1 9 0.32 12 0.27 14 0.31 12 0.28 18 0.0786 0.0712 2 9 0.32 12 0.27 14 0.31 12 0.28 14 1 35 3.87 25 0.47 35 4.60 25 0.48 21 0.0883 2 36 3.87 25 0.47 35 4.60 25 0.48 20 0.0875 1 31 1.14 9 0.15 7 0.16 7 0.16 11 0.0625 2 34 2.32 22 0.46 20 0.46 22 0.46 9 0.0351 1 3 0.14 5 0.11 3 0.13 3 0.13 35 0.2995 2 4 0.14 5 0.11 3 0.13 3 0.13 34 0.2921 1 15 0.35 16 0.30 16 0.32 16 0.31 19 0.0818 2 15 0.35 16 0.30 16 0.32 16 0.31 17 0.0766 1 27 0.78 3 0.03 1 0.03 1 0.03 29 0.1308 2 26 0.78 3 0.03 1 0.03 1 0.03 28 0.1272 1 13 0.34 1 0.03 27 1.07 33 0.68 3 0.0222 2 14 0.34 1 0.03 27 1.07 33 0.68 4 0.0226 1 24 0.72 23 0.47 21 0.47 23 0.47 10 0.0571 2 24 0.72 23 0.47 21 0.47 23 0.47 12 0.0650 1 2 0.01 33 0.60 25 0.91 31 0.60 36 NA 2 1 0.01 33 0.60 25 0.91 31 0.60 13 0.0685 1 33 1.70 31 0.53 33 1.96 29 0.54 27 0.1218 2 32 1.42 31 0.53 33 1.96 29 0.54 26 0.1178 1 18 0.39 20 0.38 18 0.39 20 0.38 6 0.0253 2 17 0.39 20 0.38 18 0.39 20 0.38 5 0.0252 Table 60: Tabulation of storage and RAM requirements for Class A — Left Index, with an enrollment set size of 100 000 subjects. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V RAM Actual Sub. # 155 On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 1 7 0.17 10 0.15 8 0.16 8 0.16 30 0.1132 2 8 0.18 10 0.15 8 0.16 8 0.16 29 0.1124 1 21 0.61 18 0.33 29 1.25 18 0.33 1 0.0190 2 21 0.61 18 0.33 30 1.40 18 0.33 1 0.0190 1 29 1.07 27 0.48 23 0.48 27 0.48 15 0.0630 2 30 1.07 27 0.48 23 0.48 27 0.48 14 0.0624 1 6 0.17 7 0.11 10 0.17 10 0.16 25 0.0933 2 5 0.17 7 0.11 10 0.17 10 0.16 22 0.0903 1 19 0.57 29 0.48 5 0.15 5 0.15 24 0.0910 0.0909 2 20 0.57 29 0.48 5 0.15 5 0.15 23 1 14 0.33 14 0.29 14 0.29 14 0.29 32 0.1230 2 13 0.33 14 0.29 14 0.29 14 0.29 33 0.1249 1 23 0.69 35 0.67 31 1.54 35 0.67 5 0.0215 2 28 1.01 36 0.99 32 1.55 36 0.99 3 0.0214 1 10 0.31 12 0.26 12 0.29 12 0.27 20 0.0708 0.0643 2 9 0.31 12 0.26 12 0.29 12 0.27 16 1 36 3.87 25 0.46 35 4.60 25 0.47 18 0.0682 2 35 3.87 25 0.46 35 4.60 25 0.47 19 0.0685 1 31 1.14 9 0.15 7 0.16 7 0.15 12 0.0505 2 34 2.32 22 0.44 20 0.45 22 0.45 9 0.0295 1 3 0.14 5 0.11 3 0.13 3 0.12 36 0.2615 2 3 0.14 5 0.11 3 0.13 3 0.12 35 0.2526 1 15 0.34 16 0.29 16 0.30 16 0.30 21 0.0776 2 16 0.34 16 0.29 16 0.30 16 0.30 17 0.0675 1 26 0.76 3 0.03 1 0.03 1 0.03 31 0.1133 2 27 0.76 3 0.03 1 0.03 1 0.03 28 0.1100 1 11 0.33 1 0.03 27 1.03 33 0.66 6 0.0218 2 11 0.33 1 0.03 27 1.03 33 0.66 3 0.0214 1 24 0.71 23 0.46 21 0.46 23 0.46 10 0.0442 2 25 0.71 23 0.46 21 0.46 23 0.46 11 0.0503 1 1 0.01 33 0.57 25 0.87 31 0.57 34 0.1929 2 1 0.01 33 0.57 25 0.87 31 0.57 13 0.0562 1 33 1.56 31 0.52 33 1.79 29 0.52 26 0.0996 2 32 1.30 31 0.52 33 1.79 29 0.52 27 0.1007 1 18 0.39 20 0.38 18 0.38 20 0.38 8 0.0223 2 17 0.39 20 0.38 18 0.38 20 0.38 7 0.0222 Table 61: Tabulation of storage and RAM requirements for Class A — Right Index, with an enrollment set size of 100 000 subjects. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 156 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V RAM Actual Sub. # On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 1 8 4.87 10 4.66 8 4.78 8 4.69 26 0.0368 2 7 4.87 10 4.66 8 4.78 8 4.69 28 0.0374 1 24 18.58 18 10.30 30 40.04 18 10.31 4 0.0030 2 24 18.58 18 10.30 31 44.68 18 10.31 4 0.0030 1 30 33.40 27 14.95 23 15.02 27 14.96 15 0.0207 2 31 33.41 27 14.95 23 15.02 27 14.96 14 0.0202 1 5 4.86 5 3.21 10 4.86 10 4.78 29 0.0386 2 5 4.86 5 3.21 10 4.86 10 4.78 30 0.0412 1 23 16.38 31 15.63 6 4.75 6 4.67 31 0.0515 0.0311 2 22 16.38 31 15.63 6 4.75 6 4.67 20 1 16 9.36 16 9.19 16 9.32 16 9.19 33 0.0686 2 15 9.36 16 9.19 16 9.32 16 9.19 32 0.0684 1 21 15.83 33 15.68 25 24.61 31 15.68 8 0.0058 2 29 30.83 36 30.68 32 46.83 36 30.68 4 0.0030 1 9 8.30 12 8.07 12 8.68 12 8.15 10 0.0143 0.0143 2 10 8.30 12 8.07 12 8.68 12 8.15 10 1 36 61.81 25 14.74 35 73.67 25 14.80 24 0.0360 2 35 61.81 25 14.74 35 73.67 25 14.80 19 0.0286 1 26 18.76 9 4.55 5 4.68 5 4.61 12 0.0146 2 34 53.98 22 13.64 20 13.73 22 13.66 9 0.0072 1 3 3.75 5 3.21 3 3.75 3 3.68 35 NA 2 4 3.76 5 3.21 3 3.75 3 3.68 34 NA 1 12 9.05 14 8.82 14 8.84 14 8.90 17 0.0229 2 11 9.05 14 8.82 14 8.84 14 8.90 16 0.0214 1 28 21.42 3 0.86 1 0.89 1 0.83 27 0.0370 2 27 21.41 3 0.86 1 0.89 1 0.83 21 0.0333 1 14 9.14 1 0.82 28 30.63 34 20.48 1 0.0027 2 13 9.14 1 0.82 28 30.63 34 20.48 1 0.0027 1 19 14.82 23 14.57 21 14.63 23 14.57 18 0.0281 2 19 14.82 23 14.57 21 14.63 23 14.57 13 0.0195 1 1 0.01 34 17.99 26 27.40 32 18.02 36 NA 2 1 0.01 34 17.99 26 27.40 32 18.02 25 0.0366 1 33 44.63 29 15.30 33 51.84 29 15.46 22 0.0336 2 32 37.35 29 15.30 33 51.84 29 15.46 23 0.0358 1 18 11.77 20 11.69 18 11.77 20 11.71 7 0.0034 2 17 11.77 20 11.69 18 11.77 20 11.71 3 0.0028 Table 62: Tabulation of storage and RAM requirements for Class A — Left and Right Index, with an enrollment set size of 1 600 000 subjects. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V RAM Actual Sub. # 3 46.49 2 4 1 12 2 13 1 On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 2 45.98 30 NA 46.34 2 45.98 29 NA 86.78 18 86.59 6 0.0020 85.95 17 85.76 2 0.0012 1 30.00 1 29.89 16 0.0043 176.65 26 176.79 28 176.67 7 0.0024 11 53.02 12 77.01 13 76.88 27 0.0591 11 53.02 12 77.01 13 76.88 27 0.0591 0.0062 4 45.92 46.49 4 79.43 16 79.43 15 14 79.73 2 28 1 2 1 157 2 46.34 45.92 2 84.51 18 83.67 15 3 29.86 317.95 30 10 77.02 10 77.02 1 25 156.12 26 152.16 4 47.21 4 47.07 18 2 26 156.12 26 152.16 4 47.21 4 47.07 14 0.0040 1 15 86.46 17 84.85 16 86.42 15 84.85 23 0.0203 2 16 86.46 17 84.85 16 86.42 15 84.85 24 0.0204 1 5 49.68 8 49.40 14 82.58 8 49.40 2 0.0012 2 21 108.71 23 108.43 23 133.80 23 108.43 1 0.0009 0.0049 1 18 101.00 21 101.67 21 100.73 21 100.75 17 2 17 101.00 21 101.67 21 100.73 21 100.75 11 0.0033 1 22 119.79 7 47.80 7 47.92 7 47.77 10 0.0031 2 27 177.48 6 46.73 6 47.56 6 47.42 11 0.0033 1 6 57.53 9 53.02 8 57.52 9 57.39 26 0.0543 2 6 57.53 9 53.02 8 57.52 9 57.39 25 0.0515 1 19 101.00 19 101.67 19 100.73 19 100.75 15 0.0041 2 20 101.00 19 101.67 19 100.73 19 100.75 13 0.0035 1 1 7.54 1 8.87 27 232.47 29 222.31 2 0.0012 2 2 7.54 1 8.87 27 232.47 29 222.31 2 0.0012 1 24 150.35 24 147.31 24 147.42 24 147.31 20 0.0108 2 23 150.35 24 147.31 24 147.42 24 147.31 21 0.0136 1 29 440.68 28 155.98 29 512.65 26 157.35 19 0.0099 2 30 540.60 29 170.16 30 625.07 27 171.53 22 0.0141 1 8 63.53 13 63.39 10 63.52 11 63.41 9 0.0027 2 9 63.53 13 63.39 10 63.52 11 63.41 7 0.0024 Table 63: Tabulation of storage and RAM requirements for Class B — Identification Flats, with an enrollment set size of 3 000 000 subjects. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 13 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 158 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V RAM Actual Sub. # On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 1 9 92.86 11 96.24 10 92.63 11 96.33 30 NA 2 10 92.86 11 96.24 10 92.63 11 96.33 24 0.0711 1 16 132.37 19 155.88 19 159.68 19 159.37 6 0.0015 2 15 132.37 18 153.13 18 156.93 18 156.62 2 0.0011 1 14 127.39 3 57.47 1 57.71 1 57.52 14 0.0088 2 28 573.25 30 318.66 26 318.89 28 318.70 13 0.0048 1 7 90.93 6 82.27 6 90.92 7 90.70 25 0.0734 2 6 90.93 6 82.27 6 90.92 7 90.70 25 0.0734 0.0368 1 25 274.35 26 267.49 4 82.40 4 82.17 23 2 26 274.35 26 267.49 4 82.40 4 82.17 20 0.0276 1 17 144.02 16 142.06 14 143.97 16 142.06 20 0.0276 2 18 144.02 16 142.06 14 143.97 16 142.06 19 0.0275 1 4 84.42 10 83.96 16 144.38 6 83.96 4 0.0013 2 11 110.22 13 109.76 17 153.35 13 109.76 1 0.0010 0.0047 1 20 181.00 22 182.11 20 182.47 22 180.58 12 2 21 181.00 23 182.11 23 182.47 23 180.58 10 0.0027 1 23 192.19 4 72.47 2 72.67 2 72.43 16 0.0102 2 27 274.95 4 72.47 3 72.67 2 72.43 15 0.0095 1 5 90.93 8 82.27 8 90.92 9 90.70 28 0.0934 2 8 90.93 8 82.27 8 90.92 9 90.70 27 0.0826 1 19 181.00 20 182.11 21 182.47 20 180.58 9 0.0025 2 22 181.00 20 182.11 21 182.47 20 180.58 10 0.0027 1 2 13.33 1 15.77 27 412.86 29 395.06 2 0.0011 2 1 13.33 1 15.77 27 412.86 29 395.06 4 0.0013 1 24 260.37 24 257.08 25 257.27 24 257.09 22 0.0311 2 3 61.04 24 257.08 24 257.27 24 257.09 29 0.1680 1 29 811.45 28 293.25 29 941.34 26 295.54 18 0.0163 2 29 811.45 28 293.25 29 941.34 26 295.54 17 0.0155 1 13 121.80 14 121.56 12 121.79 14 121.60 7 0.0024 2 12 121.79 14 121.56 12 121.79 14 121.60 7 0.0024 Table 64: Tabulation of storage and RAM requirements for Class C — Ten-Finger Plain-to-Plain, with an enrollment set size of 5 000 000 subjects. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 14 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V RAM Actual Sub. # 159 On Disk Reported Finalized Stored FNIR @ FPIR = 10−3 1 14 183.36 14 190.50 14 183.11 14 190.59 16 0.0094 2 15 183.36 14 190.50 14 183.11 14 190.59 15 0.0085 1 10 132.40 22 302.53 18 303.52 24 303.21 4 0.0015 2 9 132.37 27 309.65 19 310.65 25 310.33 7 0.0018 1 16 191.66 3 95.45 1 95.69 1 95.50 18 0.0106 2 30 930.48 30 528.40 26 528.63 28 528.45 12 0.0050 1 7 130.29 5 118.45 4 130.29 5 130.07 25 0.0536 2 6 130.29 5 118.45 4 130.29 5 130.07 25 0.0536 0.0447 1 20 261.29 18 254.66 2 97.99 2 97.76 24 2 19 261.29 18 254.66 2 97.99 2 97.76 21 0.0333 1 13 144.02 10 143.46 8 143.97 10 143.46 20 0.0201 2 12 144.02 10 143.46 8 143.97 10 143.46 19 0.0199 1 4 113.80 4 113.34 12 153.41 4 113.34 1 0.0013 2 11 137.03 9 136.57 13 153.69 9 136.57 2 0.0014 0.0051 1 22 303.13 24 304.24 21 355.15 21 302.68 13 2 24 303.13 23 304.24 20 355.15 20 302.68 9 0.0033 1 21 280.49 12 151.41 10 151.61 12 151.37 17 0.0097 2 26 367.82 12 151.41 11 151.61 12 151.37 14 0.0083 1 8 130.29 7 118.45 6 130.29 7 130.07 28 0.0783 2 5 130.29 7 118.45 6 130.29 7 130.07 27 0.0716 1 25 303.13 25 304.24 22 355.15 22 302.68 11 0.0034 2 23 303.13 25 304.24 22 355.15 22 302.68 9 0.0033 1 1 20.22 1 25.61 27 666.73 29 641.13 5 0.0017 2 2 20.22 1 25.61 27 666.73 29 641.13 2 0.0014 1 27 382.88 28 381.17 25 381.36 26 381.17 29 0.0860 2 3 78.28 28 381.17 24 381.36 26 381.17 30 0.2462 1 28 806.17 20 257.87 29 937.76 18 260.17 23 0.0358 2 28 806.17 20 257.87 29 937.76 18 260.17 22 0.0351 1 17 234.03 16 233.80 16 234.03 16 233.84 5 0.0017 2 18 234.03 16 233.80 16 234.03 16 233.84 8 0.0019 Table 65: Tabulation of storage and RAM requirements for Class C — Ten-Finger Rolled-to-Rolled, with an enrollment set size of 5 000 000 subjects. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make during the final submission period. All values are reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. Actual RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Reported RAM is the sum of the predicted RAM consumption returned from the enrollment method when enrolling images for the enrollment set. Finalized Size is the amount of disk storage space used in the finalized enrollment set directory after the finalization method returned. Stored Size is the actual amount of storage space used to hold the individual enrollment templates on disk, determined by summing the number of bytes written by the FpVTE test driver after each enrollment. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 15 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 160 G G.1 FPVTE – F INGERPRINT M ATCHING Search Template Sizes Mean Values C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Left Index Size Sub. # 161 Right Index FNIR Size Left and Right Index Size FNIR FNIR 1 5 1.49 30 0.1335 5 1.51 30 0.1132 5 2.98 26 0.0368 2 5 1.49 31 0.1337 5 1.51 29 0.1124 5 2.98 28 0.0374 1 18 3.31 1 0.0197 18 3.34 1 0.0190 18 6.65 4 0.0030 2 18 3.31 1 0.0197 18 3.34 1 0.0190 18 6.65 4 0.0030 1 27 4.83 16 0.0745 27 4.89 15 0.0630 27 9.71 15 0.0207 2 27 4.83 15 0.0723 27 4.89 14 0.0624 27 9.71 14 0.0202 1 8 1.54 25 0.1111 8 1.55 25 0.0933 7 3.00 29 0.0386 2 8 1.54 22 0.1082 8 1.55 22 0.0903 7 3.00 30 0.0412 1 10 2.51 24 0.1089 10 2.52 24 0.0910 10 5.02 31 0.0515 2 10 2.51 23 0.1086 10 2.52 23 0.0909 10 5.02 20 0.0311 1 16 3.00 32 0.1576 16 3.00 32 0.1230 16 5.99 33 0.0686 2 16 3.00 33 0.1607 16 3.00 33 0.1249 16 5.99 32 0.0684 1 35 6.83 7 0.0257 35 6.87 5 0.0215 29 10.19 8 0.0058 2 36 10.07 8 0.0278 36 10.18 3 0.0214 36 20.17 4 0.0030 1 12 2.68 18 0.0786 12 2.72 20 0.0708 12 5.26 10 0.0143 2 12 2.68 14 0.0712 12 2.72 16 0.0643 12 5.26 10 0.0143 1 25 4.82 21 0.0883 25 4.85 18 0.0682 25 9.67 24 0.0360 2 25 4.82 20 0.0875 25 4.85 19 0.0685 25 9.67 19 0.0286 1 7 1.52 11 0.0625 7 1.53 12 0.0505 9 3.01 12 0.0146 2 22 4.47 9 0.0351 22 4.51 9 0.0295 22 8.93 9 0.0072 1 3 1.21 35 0.2995 3 1.18 36 0.2615 3 2.31 35 NA 2 3 1.21 34 0.2921 3 1.18 35 0.2526 3 2.31 34 NA 1 14 2.93 19 0.0818 14 2.98 21 0.0776 14 5.77 17 0.0229 2 14 2.93 17 0.0766 14 2.98 17 0.0675 14 5.77 16 0.0214 1 1 0.28 29 0.1308 1 0.28 31 0.1133 1 0.53 27 0.0370 2 1 0.28 28 0.1272 1 0.28 28 0.1100 1 0.53 21 0.0333 1 33 6.52 3 0.0222 33 6.57 6 0.0218 34 13.08 1 0.0027 2 33 6.52 4 0.0226 33 6.57 3 0.0214 34 13.08 1 0.0027 1 23 4.74 10 0.0571 23 4.77 10 0.0442 23 9.52 18 0.0281 2 23 4.74 12 0.0650 23 4.77 11 0.0503 23 9.52 13 0.0195 1 31 6.00 36 NA 31 6.09 34 0.1929 32 12.09 36 NA 2 31 6.00 13 0.0685 31 6.09 13 0.0562 32 12.09 25 0.0366 1 29 5.17 27 0.1218 29 5.20 26 0.0996 30 10.36 22 0.0336 2 29 5.17 26 0.1178 29 5.20 27 0.1007 30 10.36 23 0.0358 1 20 3.70 6 0.0253 20 3.72 8 0.0223 20 7.41 7 0.0034 2 20 3.70 5 0.0252 20 3.72 7 0.0222 20 7.41 3 0.0028 Table 66: Tabulation of mean search template sizes for Class A. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the mean kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 162 24 0.0072 0.0392 0.0403 16 4 4 22.39 13.24 13.24 6 29 30 0.0031 NA NA 18 2 2 30.53 16.13 16.13 6 29 30 0.0020 NA NA Identification Flats Size FNIR 6.70 6 23 Left and Right Slap Size FNIR 6.70 FNIR 4 11.31 Right Slap 0.0654 4 Size 22 16 FNIR 6.59 0.0647 Left Slap 4 0.0163 Size 1 6 21 0.0024 6.59 0.0591 11.27 7 0.0591 4 27 0.0062 16 62.06 27 0.0040 2 26.80 18 0.0203 1 28 26.80 14 0.0012 13 26.45 23 0.0043 0.0049 13 26.45 2 0.0910 11 29.66 16 10 0.0901 11 30.22 28 0.0106 15 10.48 46.65 26 0.0084 1 19.21 18 0.0349 17 28 19.21 17 0.0204 0.0024 11 20.53 23 24 0.0012 0.0063 0.0083 11 20.53 2 5 0.1222 13 23.77 29.66 1 15 7 0.1220 13 15 17.69 7.78 29 0.0212 17 6 22.24 28 0.0198 0.0361 23 0.0049 0.0009 1 9.73 23.53 18 0.0641 24 0.0022 17 15 28 9.73 16 3 38.41 0.0052 11 10.32 25 23.77 1 35.39 0.0151 0.0187 11 10.32 17 14.37 21 2 0.1684 13 11.90 8 0.0068 0.0015 13 7 0.1681 13 0.0647 23 16 3.93 29 0.0371 17 26 0.0058 28.60 11.24 28 0.0325 5 25.51 1 9.66 23.22 18 0.0998 11.90 1 21 15 28 9.66 17 17 8.97 0.0156 0.0045 0.0142 11 10.25 23 10 14 0.0259 2 11 10.25 0.1008 23 14.42 5 1 13 11.92 24 0.0116 12.94 13 2 13 4 19 3.88 1 17 11.92 1 0.0287 0.0094 11.18 2 17 8.92 15 1 1 10 14.25 15 2 23 12.76 2 1 21 27 14 0.0057 0.0882 0.0904 0.0062 29 19 19 7 7 4 51.62 78.12 78.12 35.39 35.39 19.93 19.93 16.79 19 21 20 2 2 13 15 25 26 11 0.0027 0.0141 0.0099 0.0136 0.0108 0.0012 0.0012 0.0035 0.0041 0.0515 0.0543 0.0033 1 12.03 25 0.0051 29 51.62 22 0.0024 0.0033 14.11 13 0.0021 24 55.81 9 11 2 14.11 11 0.0022 24 60.97 7 21 6 25.51 2 0.0160 26 22.27 9 0.0202 6 25.51 3 0.0190 27 22.27 21 0.1259 19 58.19 21 0.0139 9 10 17 0.1155 19 58.19 22 0.0124 9 19 30 0.0142 29 40.83 20 0.0036 10 6.08 27 0.0132 29 40.83 19 0.0036 21 7.13 12 0.0057 24 41.34 7 0.0031 2 7.13 11 0.0057 24 42.94 7 10 6 12.94 3 0.0369 26 17.22 35.39 0.0276 6 12.94 3 0.0381 27 17.22 16.92 0.1736 21 29.30 21 0.0266 9 5 14 0.1634 21 29.30 22 0.0273 9 0.0047 30 0.0257 29 20.56 19 0.0106 0.0054 6.02 27 0.0254 29 20.56 20 0.0110 12 7.08 12 0.0098 24 20.73 8 25.51 2 7.08 11 0.0099 24 21.60 9 12.15 1 6 12.76 2 0.1089 26 8.65 3 2 6 12.76 3 0.1133 27 8.65 0.0126 1 19 29.00 25 0.0500 8 0.0167 2 19 29.00 26 0.0461 8 15 1 29 20.36 20 0.0192 6.14 2 29 20.36 19 0.0190 12.94 1 24 20.70 9 3 2 24 21.43 8 0.0236 1 26 8.61 0.0288 2 27 8.61 16 1 8 6.08 2 8 12.76 1 3 2 2 2 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 67: Tabulation of mean search template sizes for Class B. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the mean kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Ten-Finger Plain-to-Plain Size FNIR Sub. # 163 Ten-Finger Rolled-to-Rolled Size FNIR 9 20.91 30 NA 14 40.15 16 0.0094 2 9 20.91 24 0.0711 14 40.15 15 0.0085 1 19 33.66 6 0.0015 24 63.63 4 0.0015 2 18 33.16 2 0.0011 25 65.36 7 0.0018 1 1 12.10 14 0.0088 1 19.95 18 0.0106 2 28 67.07 13 0.0048 28 110.22 12 0.0050 1 5 19.45 25 0.0734 3 27.41 25 0.0536 2 5 19.45 25 0.0734 3 27.41 25 0.0536 1 14 27.52 23 0.0368 10 30.76 24 0.0447 2 14 27.52 20 0.0276 10 30.76 21 0.0333 1 16 29.72 20 0.0276 8 29.96 20 0.0201 2 16 29.72 19 0.0275 8 29.96 19 0.0199 1 4 17.76 4 0.0013 2 23.76 1 0.0013 2 11 23.15 1 0.0010 7 28.49 2 0.0014 1 23 38.56 12 0.0047 23 63.60 13 0.0051 2 22 38.56 10 0.0027 20 63.60 9 0.0033 1 2 15.74 16 0.0102 12 32.38 17 0.0097 2 2 15.74 15 0.0095 12 32.38 14 0.0083 1 5 19.45 28 0.0934 3 27.41 28 0.0783 2 5 19.45 27 0.0826 3 27.41 27 0.0716 1 20 38.56 9 0.0025 21 63.60 11 0.0034 2 20 38.56 10 0.0027 21 63.60 9 0.0033 1 29 83.85 2 0.0011 29 134.42 5 0.0017 2 29 83.85 4 0.0013 29 134.42 2 0.0014 1 24 54.12 22 0.0311 26 80.04 29 0.0860 2 24 54.12 29 0.1680 26 80.04 30 0.2462 1 26 62.80 18 0.0163 18 54.62 23 0.0358 2 26 62.80 17 0.0155 18 54.62 22 0.0351 1 12 25.78 7 0.0024 16 48.74 5 0.0017 2 12 25.78 7 0.0024 16 48.74 8 0.0019 1 Table 68: Tabulation of mean search template sizes for Class C. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the mean kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 164 G.2 FPVTE – F INGERPRINT M ATCHING Median Values C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Left Index Size Sub. # 165 Right Index FNIR Size Left and Right Index Size FNIR FNIR 1 5 1.49 30 0.1335 5 1.51 30 0.1132 5 2.94 26 0.0368 2 5 1.49 31 0.1337 5 1.51 29 0.1124 5 2.94 28 0.0374 1 18 3.27 1 0.0197 18 3.30 1 0.0190 18 6.57 4 0.0030 2 18 3.27 1 0.0197 18 3.30 1 0.0190 18 6.57 4 0.0030 1 27 4.84 16 0.0745 27 4.91 15 0.0630 27 9.71 15 0.0207 2 27 4.84 15 0.0723 27 4.91 14 0.0624 27 9.71 14 0.0202 1 8 1.52 25 0.1111 8 1.53 25 0.0933 8 2.97 29 0.0386 2 8 1.52 22 0.1082 8 1.53 22 0.0903 8 2.97 30 0.0412 1 10 2.35 24 0.1089 10 2.36 24 0.0910 10 4.81 31 0.0515 2 10 2.35 23 0.1086 10 2.36 23 0.0909 10 4.81 20 0.0311 1 16 3.00 32 0.1576 16 3.00 32 0.1230 16 6.00 33 0.0686 2 16 3.00 33 0.1607 16 3.00 33 0.1249 16 6.00 32 0.0684 1 35 6.84 7 0.0257 35 6.88 5 0.0215 31 10.19 8 0.0058 2 36 10.02 8 0.0278 36 10.12 3 0.0214 36 20.05 4 0.0030 1 12 2.65 18 0.0786 12 2.68 20 0.0708 12 5.19 10 0.0143 2 12 2.65 14 0.0712 12 2.68 16 0.0643 12 5.19 10 0.0143 1 25 4.76 21 0.0883 25 4.79 18 0.0682 25 9.55 24 0.0360 2 25 4.76 20 0.0875 25 4.79 19 0.0685 25 9.55 19 0.0286 1 7 1.50 11 0.0625 7 1.51 12 0.0505 7 2.97 12 0.0146 2 22 4.41 9 0.0351 22 4.45 9 0.0295 22 8.81 9 0.0072 1 3 1.20 35 0.2995 3 1.17 36 0.2615 3 2.28 35 NA 2 3 1.20 34 0.2921 3 1.17 35 0.2526 3 2.28 34 NA 1 14 2.90 19 0.0818 14 2.94 21 0.0776 14 5.69 17 0.0229 2 14 2.90 17 0.0766 14 2.94 17 0.0675 14 5.69 16 0.0214 1 1 0.28 29 0.1308 1 0.28 31 0.1133 1 0.53 27 0.0370 2 1 0.28 28 0.1272 1 0.28 28 0.1100 1 0.53 21 0.0333 1 33 6.46 3 0.0222 33 6.51 6 0.0218 34 12.94 1 0.0027 2 33 6.46 4 0.0226 33 6.51 3 0.0214 34 12.94 1 0.0027 1 23 4.69 10 0.0571 23 4.71 10 0.0442 23 9.40 18 0.0281 2 23 4.69 12 0.0650 23 4.71 11 0.0503 23 9.40 13 0.0195 1 31 5.81 36 NA 31 5.89 34 0.1929 32 11.73 36 NA 2 31 5.81 13 0.0685 31 5.89 13 0.0562 32 11.73 25 0.0366 1 29 5.01 27 0.1218 29 5.04 26 0.0996 29 10.05 22 0.0336 2 29 5.01 26 0.1178 29 5.04 27 0.1007 29 10.05 23 0.0358 1 20 3.63 6 0.0253 20 3.65 8 0.0223 20 7.27 7 0.0034 2 20 3.63 5 0.0252 20 3.65 7 0.0222 20 7.27 3 0.0028 Table 69: Tabulation of median search template sizes for Class A. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the median kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 166 24 0.0072 0.0392 0.0403 16 4 4 22.48 13.18 13.18 6 29 30 0.0031 NA NA 18 2 2 30.57 16.03 16.03 6 29 30 0.0020 NA NA Identification Flats Size FNIR 6.67 6 23 Left and Right Slap Size FNIR 6.67 FNIR 4 11.35 Right Slap 0.0654 4 Size 22 16 FNIR 6.56 0.0647 Left Slap 4 0.0163 Size 1 6 21 0.0024 6.56 0.0591 11.30 7 0.0591 4 27 0.0062 16 62.33 27 0.0040 2 26.48 18 0.0203 1 28 26.48 14 0.0012 13 25.93 23 0.0043 0.0049 13 25.93 2 0.0910 11 30.00 16 10 0.0901 11 30.17 28 0.0106 15 10.43 46.93 26 0.0084 1 18.78 18 0.0349 17 28 18.78 17 0.0204 0.0024 11 20.12 23 24 0.0012 0.0063 0.0083 11 20.12 2 5 0.1222 13 24.00 30.00 1 15 7 0.1220 13 15 17.87 7.78 29 0.0212 17 6 22.24 28 0.0198 0.0361 23 0.0049 0.0009 1 9.49 23.68 18 0.0641 24 0.0022 17 15 28 9.49 16 3 38.44 0.0052 11 10.07 25 24.00 1 35.37 0.0151 0.0187 11 10.07 17 14.52 19 2 0.1684 13 12.00 8 0.0068 0.0015 13 7 0.1681 13 0.0647 23 16 3.91 29 0.0371 17 26 0.0058 28.63 11.24 28 0.0325 5 25.46 1 9.41 23.35 18 0.0998 12.00 1 19 15 28 9.41 17 17 9.05 0.0156 0.0045 0.0142 11 9.99 23 10 14 0.0259 2 11 9.99 0.1008 23 14.42 5 1 13 12.00 24 0.0116 12.93 13 2 13 4 19 3.89 1 17 12.00 1 0.0287 0.0094 11.18 2 17 9.00 15 1 1 10 14.26 15 2 23 12.73 2 1 19 27 14 0.0057 0.0882 0.0904 0.0062 29 19 19 7 7 4 51.27 78.05 78.05 35.37 35.37 19.73 19.73 16.74 19 21 20 2 2 13 15 25 26 11 0.0027 0.0141 0.0099 0.0136 0.0108 0.0012 0.0012 0.0035 0.0041 0.0515 0.0543 0.0033 1 12.00 25 0.0051 29 51.27 22 0.0024 0.0033 13.82 13 0.0021 24 56.14 9 11 2 13.82 11 0.0022 24 60.55 7 19 6 25.46 2 0.0160 26 22.46 9 0.0202 6 25.46 3 0.0190 27 22.46 19 0.1259 19 58.20 21 0.0139 9 10 17 0.1155 19 58.20 22 0.0124 9 19 30 0.0142 29 40.31 20 0.0036 10 6.06 27 0.0132 29 40.31 19 0.0036 19 6.97 12 0.0057 24 41.58 7 0.0031 2 6.97 11 0.0057 24 42.60 7 10 6 12.93 3 0.0369 26 17.26 35.37 0.0276 6 12.93 3 0.0381 27 17.26 16.86 0.1736 19 29.30 21 0.0266 9 5 14 0.1634 19 29.30 22 0.0273 9 0.0047 30 0.0257 29 20.25 19 0.0106 0.0054 6.00 27 0.0254 29 20.25 20 0.0110 12 6.91 12 0.0098 24 20.92 8 25.46 2 6.91 11 0.0099 24 21.43 9 12.12 1 6 12.73 2 0.1089 26 8.66 3 2 6 12.73 3 0.1133 27 8.66 0.0126 1 19 29.00 25 0.0500 8 0.0167 2 19 29.00 26 0.0461 8 15 1 29 20.07 20 0.0192 6.11 2 29 20.07 19 0.0190 12.93 1 24 20.84 9 3 2 24 21.27 8 0.0236 1 26 8.62 0.0288 2 27 8.62 16 1 8 6.06 2 8 12.73 1 3 2 2 2 1 Sub. # Participant Letter C D E F G H I J L M O Q S U V Table 70: Tabulation of median search template sizes for Class B. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the median kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Ten-Finger Plain-to-Plain Size FNIR Sub. # 167 Ten-Finger Rolled-to-Rolled Size FNIR 9 20.80 30 NA 14 40.13 16 0.0094 2 9 20.80 24 0.0711 14 40.13 15 0.0085 1 19 33.58 6 0.0015 20 63.49 4 0.0015 2 18 33.00 2 0.0011 25 65.19 7 0.0018 1 1 12.07 14 0.0088 1 19.69 18 0.0106 2 28 66.94 13 0.0048 28 108.80 12 0.0050 1 5 19.35 25 0.0734 3 27.48 25 0.0536 2 5 19.35 25 0.0734 3 27.48 25 0.0536 1 14 26.87 23 0.0368 10 30.12 24 0.0447 2 14 26.87 20 0.0276 10 30.12 21 0.0333 1 16 30.00 20 0.0276 8 30.00 20 0.0201 2 16 30.00 19 0.0275 8 30.00 19 0.0199 1 4 17.86 4 0.0013 2 23.92 1 0.0013 2 11 23.17 1 0.0010 7 28.90 2 0.0014 1 20 38.52 12 0.0047 21 63.55 13 0.0051 2 20 38.52 10 0.0027 21 63.55 9 0.0033 1 2 15.77 16 0.0102 12 32.65 17 0.0097 2 2 15.77 15 0.0095 12 32.65 14 0.0083 1 5 19.35 28 0.0934 3 27.48 28 0.0783 2 5 19.35 27 0.0826 3 27.48 27 0.0716 1 20 38.52 9 0.0025 21 63.55 11 0.0034 2 20 38.52 10 0.0027 21 63.55 9 0.0033 1 29 83.49 2 0.0011 29 135.01 5 0.0017 2 29 83.49 4 0.0013 29 135.01 2 0.0014 1 24 53.59 22 0.0311 26 78.67 29 0.0860 2 24 53.59 29 0.1680 26 78.67 30 0.2462 1 26 63.23 18 0.0163 18 54.73 23 0.0358 2 26 63.23 17 0.0155 18 54.73 22 0.0351 1 12 25.68 7 0.0024 16 48.85 5 0.0017 2 12 25.68 7 0.0024 16 48.85 8 0.0019 1 Table 71: Tabulation of median search template sizes for Class C. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Size values indicate the median kilobytes used to store a search template on disk for a single subject, where 1 kB is equal to 1 024 bytes. The FNIR column shows FNIR for each submission at FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 168 H FPVTE – F INGERPRINT M ATCHING Template Creation Times C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Enrollment Mean Sub. # 169 Search Median Mean Median FNIR @ FPIR = 10−3 1 21 0.45 19 0.41 19 0.41 19 0.39 30 0.1335 2 21 0.45 19 0.41 19 0.41 19 0.39 31 0.1337 1 35 1.79 34 1.46 35 1.57 34 1.43 1 0.0197 0.0197 2 34 1.79 33 1.46 34 1.55 33 1.42 1 1 11 0.37 15 0.36 15 0.35 15 0.35 16 0.0745 2 11 0.37 15 0.36 15 0.35 15 0.35 15 0.0723 1 31 1.21 31 1.13 29 1.11 29 1.07 25 0.1111 2 31 1.21 31 1.13 29 1.11 29 1.07 22 0.1082 1 19 0.43 21 0.41 21 0.57 23 0.55 24 0.1089 0.1086 2 19 0.43 21 0.41 21 0.57 23 0.55 23 1 13 0.37 7 0.31 11 0.33 7 0.30 32 0.1576 2 13 0.37 7 0.31 11 0.33 7 0.30 33 0.1607 1 33 1.49 35 1.47 33 1.48 35 1.46 7 0.0257 2 36 2.29 36 2.20 36 2.19 36 2.15 8 0.0278 1 7 0.35 13 0.34 9 0.33 13 0.32 18 0.0786 2 7 0.35 13 0.34 9 0.33 13 0.32 14 0.0712 1 27 1.09 27 1.07 27 1.06 27 1.05 21 0.0883 2 27 1.09 27 1.07 27 1.06 27 1.05 20 0.0875 1 1 0.07 1 0.06 1 0.07 1 0.05 11 0.0625 2 4 0.19 4 0.16 4 0.17 4 0.15 9 0.0351 1 29 1.21 29 1.13 31 1.12 31 1.07 35 0.2995 2 29 1.21 29 1.13 31 1.12 31 1.07 34 0.2921 1 5 0.29 5 0.28 5 0.28 5 0.27 19 0.0818 2 5 0.29 5 0.28 5 0.28 5 0.27 17 0.0766 1 17 0.42 17 0.38 17 0.38 17 0.36 29 0.1308 2 17 0.42 17 0.38 17 0.38 17 0.36 28 0.1272 1 23 0.64 23 0.56 23 0.57 21 0.54 3 0.0222 2 23 0.64 23 0.56 23 0.57 21 0.54 4 0.0226 0.0571 1 25 0.91 25 0.89 25 0.90 25 0.88 10 2 25 0.91 25 0.89 25 0.90 25 0.88 12 0.0650 1 9 0.35 9 0.32 7 0.33 9 0.31 36 NA 2 9 0.35 9 0.32 7 0.33 9 0.31 13 0.0685 1 2 0.17 2 0.16 2 0.16 2 0.15 27 0.1218 2 2 0.17 2 0.16 2 0.16 2 0.15 26 0.1178 0.32 6 0.0253 0.32 5 0.0252 1 15 2 15 0.37 11 0.37 11 0.33 13 0.33 13 0.34 11 0.34 11 Table 72: Tabulation of enrollment time results for Class A — Left Index. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 7 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 170 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Enrollment Mean Sub. # Search Median Mean Median FNIR @ FPIR = 10−3 1 21 0.45 19 0.41 19 0.41 19 0.39 30 0.1132 2 21 0.45 19 0.41 19 0.41 19 0.39 29 0.1124 1 34 1.78 33 1.45 35 1.58 34 1.43 1 0.0190 0.0190 2 35 1.80 34 1.46 34 1.57 33 1.43 1 1 13 0.37 15 0.37 15 0.37 15 0.36 15 0.0630 2 13 0.37 15 0.37 15 0.37 15 0.36 14 0.0624 1 29 1.23 29 1.15 31 1.14 29 1.09 25 0.0933 2 29 1.23 29 1.15 31 1.14 29 1.09 22 0.0903 1 19 0.43 21 0.41 21 0.57 23 0.55 24 0.0910 0.0909 2 19 0.43 21 0.41 21 0.57 23 0.55 23 1 15 0.38 7 0.31 9 0.34 7 0.30 32 0.1230 2 15 0.38 7 0.31 9 0.34 7 0.30 33 0.1249 1 33 1.51 35 1.48 33 1.49 35 1.46 5 0.0215 2 36 2.29 36 2.20 36 2.19 36 2.15 3 0.0214 1 7 0.35 13 0.33 11 0.34 13 0.33 20 0.0708 2 7 0.35 13 0.33 11 0.34 13 0.33 16 0.0643 1 27 1.09 27 1.06 27 1.06 27 1.05 18 0.0682 2 27 1.09 27 1.06 27 1.06 27 1.05 19 0.0685 1 1 0.07 1 0.06 1 0.07 1 0.05 12 0.0505 2 4 0.19 4 0.16 4 0.17 4 0.16 9 0.0295 1 31 1.24 31 1.16 29 1.14 31 1.09 36 0.2615 2 31 1.24 31 1.16 29 1.14 31 1.09 35 0.2526 1 5 0.29 5 0.28 5 0.28 5 0.27 21 0.0776 2 5 0.29 5 0.28 5 0.28 5 0.27 17 0.0675 1 17 0.42 17 0.38 17 0.39 17 0.37 31 0.1133 2 17 0.42 17 0.38 17 0.39 17 0.37 28 0.1100 1 23 0.64 23 0.56 23 0.58 21 0.54 6 0.0218 2 23 0.64 23 0.56 23 0.58 21 0.54 3 0.0214 0.0442 1 25 0.91 25 0.89 25 0.89 25 0.88 10 2 25 0.91 25 0.89 25 0.89 25 0.88 11 0.0503 1 9 0.35 9 0.32 7 0.33 9 0.31 34 0.1929 2 9 0.35 9 0.32 7 0.33 9 0.31 13 0.0562 1 2 0.17 2 0.16 2 0.16 2 0.15 26 0.0996 2 2 0.17 2 0.16 2 0.16 2 0.15 27 0.1007 0.32 8 0.0223 0.32 7 0.0222 1 11 2 11 0.37 11 0.37 11 0.33 13 0.33 13 0.34 11 0.34 11 Table 73: Tabulation of enrollment time results for Class A — Right Index. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 8 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J K L M O P Q S T U V Enrollment Mean Sub. # 171 Search Median Mean Median FNIR @ FPIR = 10−3 1 21 0.88 21 0.80 19 0.80 19 0.78 26 0.0368 2 21 0.88 21 0.80 19 0.80 19 0.78 28 0.0374 1 35 3.57 34 2.91 34 3.10 33 2.84 4 0.0030 0.0030 2 34 3.55 33 2.89 35 3.12 34 2.84 4 1 15 0.74 15 0.73 15 0.70 15 0.70 15 0.0207 2 15 0.74 15 0.73 15 0.70 15 0.70 14 0.0202 1 29 2.42 29 2.26 29 2.23 29 2.15 29 0.0386 2 29 2.42 29 2.26 29 2.23 29 2.15 30 0.0412 1 19 0.84 19 0.80 21 1.13 23 1.09 31 0.0515 0.0311 2 19 0.84 19 0.80 21 1.13 23 1.09 20 1 13 0.74 7 0.62 9 0.66 7 0.60 33 0.0686 2 13 0.74 7 0.62 9 0.66 7 0.60 32 0.0684 1 33 2.98 35 2.96 33 2.96 35 2.93 8 0.0058 2 36 4.59 36 4.40 36 4.37 36 4.30 4 0.0030 1 7 0.69 13 0.67 11 0.66 13 0.65 10 0.0143 2 7 0.69 13 0.67 11 0.66 13 0.65 10 0.0143 1 27 2.17 27 2.13 27 2.11 27 2.09 24 0.0360 2 27 2.17 27 2.13 27 2.11 27 2.09 19 0.0286 1 1 0.12 1 0.11 1 0.11 1 0.11 12 0.0146 2 4 0.36 4 0.32 4 0.33 4 0.31 9 0.0072 1 31 2.42 31 2.27 31 2.25 31 2.16 35 NA 2 31 2.42 31 2.27 31 2.25 31 2.16 34 NA 1 5 0.57 5 0.55 5 0.55 5 0.53 17 0.0229 2 5 0.57 5 0.55 5 0.55 5 0.53 16 0.0214 1 17 0.83 17 0.76 17 0.77 17 0.72 27 0.0370 2 17 0.83 17 0.76 17 0.77 17 0.72 21 0.0333 1 23 1.27 23 1.10 23 1.13 21 1.08 1 0.0027 2 23 1.27 23 1.10 23 1.13 21 1.08 1 0.0027 0.0281 1 25 1.81 25 1.78 25 1.77 25 1.75 18 2 25 1.81 25 1.78 25 1.77 25 1.75 13 0.0195 1 9 0.70 9 0.64 7 0.65 9 0.62 36 NA 2 9 0.70 9 0.64 7 0.65 9 0.62 25 0.0366 1 2 0.32 2 0.31 2 0.30 2 0.30 22 0.0336 2 2 0.32 2 0.31 2 0.30 2 0.30 23 0.0358 0.64 7 0.0034 0.64 3 0.0028 1 11 2 11 0.73 11 0.73 11 0.65 13 0.65 13 0.68 11 0.68 11 Table 74: Tabulation of enrollment time results for Class A — Left and Right Index. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 9 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 172 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Search Mean Sub. # Median FNIR @ FPIR = 10−3 2.12 22 0.0654 2.13 12 2.12 21 0.0647 14 2.16 14 2.15 6 0.0163 2 24 4.18 24 4.17 5 0.0142 1 11 1.95 11 1.95 13 0.0259 2 15 2.21 15 2.21 7 0.0187 0.1684 1 12 2.13 12 2 12 1 1 28 6.94 26 6.73 29 2 28 6.94 26 6.73 28 0.1681 1 18 2.81 18 2.83 18 0.0371 2 18 2.81 18 2.83 17 0.0325 1 9 1.68 9 1.67 23 0.0998 2 9 1.68 9 1.67 24 0.1008 1 25 6.50 25 6.50 4 0.0116 2 30 7.68 30 7.67 1 0.0094 1 7 1.22 3 1.21 15 0.0287 2 7 1.22 3 1.21 10 0.0236 1 2 1.13 2 1.11 16 0.0288 2 1 0.33 1 0.33 14 0.0276 1 26 6.94 28 6.73 30 0.1736 2 26 6.94 28 6.73 27 0.1634 1 5 1.22 5 1.21 12 0.0257 2 5 1.22 5 1.21 11 0.0254 1 20 3.41 20 3.41 2 0.0098 2 20 3.41 20 3.41 3 0.0099 1 22 3.97 22 3.96 25 0.1089 0.1133 2 22 3.97 22 3.96 26 1 17 2.46 17 2.42 20 0.0500 2 16 2.45 16 2.41 19 0.0461 1 3 1.21 7 1.22 9 0.0192 2 3 1.21 7 1.22 8 0.0190 Table 75: Tabulation of enrollment time results for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 10 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Search Mean Sub. # Median 173 FNIR @ FPIR = 10−3 2.09 24 0.0403 2.10 12 2.09 23 0.0392 14 2.15 14 2.14 6 0.0072 2 24 4.15 24 4.13 2 0.0052 1 11 1.92 11 1.91 13 0.0151 2 15 2.18 15 2.18 7 0.0083 0.1222 1 12 2.10 12 2 12 1 1 26 6.86 26 6.68 29 2 26 6.86 26 6.68 28 0.1220 1 18 2.76 18 2.78 18 0.0212 2 18 2.76 18 2.78 16 0.0198 1 9 1.64 9 1.63 25 0.0641 2 9 1.64 9 1.63 26 0.0647 1 25 6.47 25 6.47 5 0.0058 2 30 7.65 30 7.65 1 0.0045 1 5 1.21 5 1.20 14 0.0156 2 5 1.21 5 1.20 10 0.0126 1 2 1.12 2 1.11 15 0.0167 2 1 0.33 1 0.33 17 0.0202 1 28 6.88 28 6.69 30 0.1259 2 28 6.88 28 6.69 27 0.1155 1 7 1.21 7 1.20 12 0.0142 2 7 1.21 7 1.20 11 0.0132 1 20 3.41 20 3.41 3 0.0057 2 20 3.41 20 3.41 3 0.0057 1 22 3.89 22 3.88 21 0.0369 0.0381 2 22 3.89 22 3.88 22 1 16 2.43 16 2.39 19 0.0266 2 17 2.43 17 2.39 20 0.0273 1 3 1.19 3 1.19 8 0.0106 2 3 1.19 3 1.19 9 0.0110 Table 76: Tabulation of enrollment time results for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 11 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 174 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Search Mean Sub. # Median FNIR @ FPIR = 10−3 4.20 30 NA 4.22 12 4.20 29 NA 14 4.29 14 4.28 6 0.0031 2 24 8.31 24 8.28 5 0.0024 1 11 3.86 11 3.85 15 0.0063 2 15 4.39 15 4.38 10 0.0049 0.0910 1 12 4.22 12 2 12 1 1 26 13.78 26 13.49 28 2 26 13.78 26 13.49 26 0.0901 1 18 5.59 18 5.62 18 0.0106 2 18 5.59 18 5.62 17 0.0084 1 9 3.31 9 3.31 23 0.0349 2 9 3.31 9 3.31 24 0.0361 1 25 12.94 25 12.96 3 0.0022 2 30 15.30 30 15.33 1 0.0015 1 7 2.42 5 2.40 16 0.0068 2 7 2.42 5 2.40 9 0.0047 1 2 2.24 2 2.23 12 0.0054 2 1 0.66 1 0.66 14 0.0062 1 28 13.79 28 13.53 27 0.0904 2 28 13.79 28 13.53 25 0.0882 1 5 2.42 3 2.40 13 0.0057 2 5 2.42 3 2.40 11 0.0051 1 20 6.84 20 6.84 2 0.0021 2 20 6.84 20 6.84 3 0.0022 1 22 7.82 22 7.83 21 0.0160 0.0190 2 22 7.82 22 7.83 22 1 17 4.91 16 4.82 20 0.0139 2 16 4.89 17 4.83 19 0.0124 1 3 2.40 7 2.40 7 0.0036 2 3 2.40 7 2.40 7 0.0036 Table 77: Tabulation of enrollment time results for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 12 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Enrollment Mean Sub. # 1 11 175 Search Median Mean Median FNIR @ FPIR = 10−3 5.13 30 NA 5.16 11 5.13 29 NA 17 6.19 17 6.16 6 0.0020 11.47 24 11.90 24 11.86 2 0.0012 13 5.28 13 5.35 13 5.34 16 0.0043 18 5.89 16 5.95 16 5.93 7 0.0024 0.0591 4.92 11 5.06 11 5.16 11 2 11 4.92 11 5.06 11 1 19 5.82 19 5.97 2 24 11.20 24 1 13 5.11 2 18 5.69 1 28 17.69 28 17.86 28 18.36 26 18.07 27 2 28 17.69 28 17.86 28 18.36 26 18.07 27 0.0591 1 14 5.41 14 5.60 18 7.89 18 7.90 18 0.0062 2 14 5.41 14 5.60 18 7.89 18 7.90 14 0.0040 1 9 4.15 9 4.28 9 4.37 9 4.37 23 0.0203 2 9 4.15 9 4.28 9 4.37 9 4.37 24 0.0204 1 25 15.80 25 16.47 25 16.50 25 16.52 2 0.0012 2 30 18.56 30 19.35 30 19.36 30 19.36 1 0.0009 1 3 3.20 3 3.27 7 3.38 3 3.35 17 0.0049 2 3 3.20 3 3.27 7 3.38 3 3.35 11 0.0033 1 2 2.94 2 3.02 2 3.05 2 3.05 10 0.0031 2 1 0.84 1 0.87 1 0.88 1 0.88 11 0.0033 1 26 17.66 26 17.85 26 18.36 28 18.11 26 0.0543 2 26 17.66 26 17.85 26 18.36 28 18.11 25 0.0515 1 7 3.20 5 3.27 5 3.38 5 3.36 15 0.0041 2 7 3.20 5 3.27 5 3.38 5 3.36 13 0.0035 1 20 8.50 20 8.85 20 8.89 20 8.92 2 0.0012 2 20 8.50 20 8.85 20 8.89 20 8.92 2 0.0012 1 22 9.80 22 10.19 22 10.24 22 10.31 20 0.0108 0.0136 2 22 9.80 22 10.19 22 10.24 22 10.31 21 1 17 5.66 17 5.73 15 5.90 15 5.82 19 0.0099 2 16 5.56 16 5.66 14 5.80 14 5.73 22 0.0141 1 5 3.20 7 3.30 3 3.35 7 3.36 9 0.0027 2 5 3.20 7 3.30 3 3.35 7 3.36 7 0.0024 Table 78: Tabulation of enrollment time results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 13 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 176 FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Enrollment Mean Sub. # 1 15 Search Median 5.93 13 2 15 1 19 2 Mean Median FNIR @ FPIR = 10−3 5.80 30 NA 5.84 13 5.80 24 0.0711 7.12 17 7.10 6 0.0015 24 13.69 24 13.66 2 0.0011 5.21 11 5.20 11 5.20 14 0.0088 5.22 12 5.21 12 5.20 13 0.0048 0.0734 5.89 13 5.84 13 5.93 13 5.89 13 7.10 19 7.09 17 24 13.55 24 13.52 1 11 5.19 11 2 12 5.20 12 1 27 15.96 25 15.88 27 15.96 27 15.87 25 2 27 15.96 25 15.88 27 15.96 27 15.87 25 0.0734 1 13 5.93 15 5.91 18 8.35 18 8.32 23 0.0368 2 13 5.93 15 5.91 18 8.35 18 8.32 20 0.0276 1 9 5.18 9 5.21 9 5.16 9 5.16 20 0.0276 2 9 5.18 9 5.21 9 5.16 9 5.16 19 0.0275 1 29 16.65 29 16.67 29 16.71 29 16.67 4 0.0013 2 30 17.15 30 17.07 30 17.04 30 16.99 1 0.0010 1 3 3.61 3 3.58 3 3.62 5 3.62 12 0.0047 2 6 3.63 6 3.60 6 3.64 6 3.62 10 0.0027 1 1 3.56 1 3.52 1 3.51 1 3.48 16 0.0102 2 1 3.56 1 3.52 1 3.51 1 3.48 15 0.0095 1 25 15.95 27 15.89 25 15.91 25 15.83 28 0.0934 2 25 15.95 27 15.89 25 15.91 25 15.83 27 0.0826 1 4 3.63 4 3.60 4 3.63 3 3.61 9 0.0025 2 4 3.63 4 3.60 4 3.63 3 3.61 10 0.0027 1 20 9.10 20 9.08 20 9.14 20 9.11 2 0.0011 2 20 9.10 20 9.08 20 9.14 20 9.11 4 0.0013 1 22 10.86 22 10.87 22 10.91 22 10.87 22 0.0311 0.1680 2 22 10.86 22 10.87 22 10.91 22 10.87 29 1 17 6.34 17 6.09 15 6.13 15 5.91 18 0.0163 2 17 6.34 17 6.09 15 6.13 15 5.91 17 0.0155 1 7 3.68 7 3.67 7 3.69 7 3.66 7 0.0024 2 7 3.68 7 3.67 7 3.69 7 3.66 7 0.0024 Table 79: Tabulation of enrollment time results for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 14 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter C D E F G H I J L M O Q S U V Enrollment Mean Sub. # 1 13 177 Search Median Mean Median FNIR @ FPIR = 10−3 10.69 16 0.0094 10.79 13 10.69 15 0.0085 23 17.44 21 17.39 4 0.0015 30.70 30 30.43 30 30.47 7 0.0018 12 9.94 12 9.90 12 9.92 18 0.0106 11 9.94 11 9.89 11 9.91 12 0.0050 0.0536 11.18 13 11.00 13 10.79 13 2 13 11.18 13 11.00 13 1 21 17.37 21 17.27 2 30 30.78 30 1 12 9.93 2 11 9.92 1 26 21.06 28 20.95 28 21.07 28 20.92 25 2 26 21.06 28 20.95 28 21.07 28 20.92 25 0.0536 1 15 11.21 15 11.19 19 15.55 19 15.50 24 0.0447 2 15 11.21 15 11.19 19 15.55 19 15.50 21 0.0333 1 17 12.36 17 12.36 15 12.08 15 12.17 20 0.0201 2 17 12.36 17 12.36 15 12.08 15 12.17 19 0.0199 1 25 20.25 25 20.25 25 20.21 25 20.23 1 0.0013 2 24 18.75 24 18.76 24 18.72 24 18.73 2 0.0014 1 7 6.78 5 6.67 5 6.78 5 6.69 13 0.0051 2 8 6.80 8 6.71 8 6.82 8 6.74 9 0.0033 1 3 4.43 3 4.44 3 4.36 3 4.36 17 0.0097 2 3 4.43 3 4.44 3 4.36 3 4.36 14 0.0083 1 28 21.08 26 20.92 26 21.02 26 20.89 28 0.0783 2 28 21.08 26 20.92 26 21.02 26 20.89 27 0.0716 1 5 6.77 6 6.67 6 6.80 6 6.71 11 0.0034 2 5 6.77 6 6.67 6 6.80 6 6.71 9 0.0033 1 19 12.94 19 13.03 17 12.95 17 13.06 5 0.0017 2 19 12.94 19 13.03 17 12.95 17 13.06 2 0.0014 1 22 17.54 22 17.67 21 17.43 22 17.57 29 0.0860 0.2462 2 22 17.54 22 17.67 21 17.43 22 17.57 30 1 1 2.95 1 2.88 1 2.94 1 2.87 23 0.0358 2 1 2.95 1 2.88 1 2.94 1 2.87 22 0.0351 1 9 8.60 9 8.62 9 8.48 9 8.50 5 0.0017 2 9 8.60 9 8.62 9 8.48 9 8.50 8 0.0019 Table 80: Tabulation of enrollment time results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Enrollment shows the time used to create a fingerprint template to be used in an enrollment set. Search shows the time used to create a search template to be used for a query. All values are reported in seconds, but were originally recorded to microsecond precision. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. For reference, the FNIR values from Table 15 are reprinted to the right of this table. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 178 I FPVTE – F INGERPRINT M ATCHING Ranked Results In order to reduce the number of tables in Section 10 of the main body of the report, this appendix contains the tables showing ranked results for all three classes. The tables from the main body of the document are repeated in this appendix so there is a complete set of tables here for the reader to analyze. Class A results are in Tables 81 through 83, Class B results are in Tables 84 through 87, and Class C results are in Tables 88 through 89. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Letter Sub. # Identification Mean Median FNIR 179 Search Enrollment Mean Median RAM D 1 1 0.0197 20 7.52 20 7.32 30 1.57 29 1.43 19 0.61 Q 1 2 0.0222 29 15.90 28 15.70 20 0.57 19 0.54 13 0.34 Q 2 3 0.0226 23 10.42 24 10.43 20 0.57 19 0.54 14 0.34 V 1 4 0.0253 19 6.22 19 6.01 12 0.34 10 0.32 17 0.39 I 1 5 0.0257 22 10.36 22 9.94 29 1.48 30 1.46 20 0.70 L 2 6 0.0351 12 3.34 12 3.32 3 0.17 3 0.15 28 2.32 S 1 7 0.0571 28 15.37 29 16.86 22 0.90 22 0.88 21 0.72 L 1 8 0.0625 2 0.30 2 0.29 1 0.07 1 0.05 26 1.14 T 2 9 0.0685 18 5.96 18 5.97 6 0.33 8 0.31 1 0.01 J 2 10 0.0712 7 1.34 7 1.25 8 0.33 11 0.32 9 0.32 E 2 11 0.0723 30 16.71 30 16.90 13 0.35 13 0.35 24 1.11 E 1 12 0.0745 4 0.57 3 0.35 13 0.35 13 0.35 25 1.12 O 2 13 0.0766 9 1.63 9 1.56 4 0.28 4 0.27 15 0.35 J 1 14 0.0786 3 0.56 4 0.54 8 0.33 11 0.32 9 0.32 O 1 15 0.0818 5 0.64 5 0.62 4 0.28 4 0.27 15 0.35 K 2 16 0.0875 24 10.47 23 10.32 23 1.06 23 1.05 30 3.87 K 1 17 0.0883 25 10.48 25 10.44 23 1.06 23 1.05 29 3.87 F 2 18 0.1082 15 3.78 14 3.56 25 1.11 25 1.07 5 0.18 G 1 19 0.1089 21 9.52 21 9.74 19 0.57 21 0.55 18 0.57 F 1 20 0.1111 11 2.52 11 2.49 25 1.11 25 1.07 6 0.18 U 1 21 0.1218 26 14.72 27 14.42 2 0.16 2 0.15 27 1.70 P 2 22 0.1272 13 3.43 13 3.33 15 0.38 15 0.36 22 0.78 P 1 23 0.1308 8 1.38 8 1.32 15 0.38 15 0.36 23 0.78 C 1 24 0.1335 1 0.29 1 0.26 17 0.41 17 0.39 7 0.18 C 2 25 0.1337 6 0.87 6 0.76 17 0.41 17 0.39 8 0.18 H 1 26 0.1576 14 3.65 15 3.61 10 0.33 6 0.30 11 0.33 H 2 27 0.1607 17 4.26 17 4.13 10 0.33 6 0.30 11 0.33 M 2 28 0.2921 27 15.19 26 10.70 27 1.12 27 1.07 4 0.14 M 1 29 0.2995 10 2.07 10 1.84 27 1.12 27 1.07 3 0.14 T 1 30 NA 16 4.18 16 3.99 6 0.33 8 0.31 2 0.01 D 2 1 0.0197 2 42.64 3 41.84 5 1.55 5 1.42 3 0.61 V 2 2 0.0252 6 66.23 6 65.35 2 0.34 2 0.32 1 0.39 I 2 3 0.0278 3 43.24 2 40.99 6 2.19 6 2.15 5 1.04 S 2 4 0.0650 4 45.59 4 44.66 4 0.90 4 0.88 4 0.72 G 2 5 0.1086 5 54.24 5 54.03 3 0.57 3 0.55 2 0.57 U 2 6 0.1178 1 25.66 1 24.16 1 0.16 1 0.15 6 1.42 Table 81: Tabulation of ranked results for Class A — Left Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 100 000, as seen in Table 17. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 72. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 60. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 180 FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Letter Sub. # Identification Mean Median FNIR Search Enrollment Mean Median RAM D 1 1 0.0190 20 7.68 20 7.46 30 1.58 29 1.43 19 0.61 Q 2 2 0.0214 22 10.15 23 9.98 20 0.58 19 0.54 11 0.33 I 1 3 0.0215 23 10.19 22 9.83 29 1.49 30 1.46 20 0.69 Q 1 4 0.0218 28 14.44 28 14.24 20 0.58 19 0.54 11 0.33 V 1 5 0.0223 19 5.95 18 5.60 12 0.34 10 0.32 17 0.39 L 2 6 0.0295 12 3.26 13 3.26 3 0.17 3 0.16 28 2.32 S 1 7 0.0442 29 15.08 30 16.55 22 0.89 22 0.88 21 0.71 L 1 8 0.0505 2 0.29 2 0.26 1 0.07 1 0.05 26 1.14 T 2 9 0.0562 18 5.64 19 5.87 6 0.33 8 0.31 1 0.01 E 2 10 0.0624 30 15.89 29 16.27 13 0.37 13 0.36 25 1.07 E 1 11 0.0630 3 0.45 3 0.32 13 0.37 13 0.36 24 1.07 J 2 12 0.0643 7 1.25 7 1.13 10 0.34 11 0.33 9 0.31 O 2 13 0.0675 9 1.48 9 1.36 4 0.28 4 0.27 16 0.34 K 1 14 0.0682 25 10.42 26 10.42 23 1.06 23 1.05 30 3.87 K 2 15 0.0685 24 10.30 25 10.27 23 1.06 23 1.05 29 3.87 J 1 16 0.0708 4 0.53 4 0.51 10 0.34 11 0.33 10 0.31 O 1 17 0.0776 5 0.59 5 0.56 4 0.28 4 0.27 15 0.34 F 2 18 0.0903 15 3.71 15 3.60 27 1.14 25 1.09 5 0.17 G 1 19 0.0910 21 10.12 24 10.10 19 0.57 21 0.55 18 0.57 F 1 20 0.0933 11 2.52 11 2.38 27 1.14 25 1.09 6 0.17 U 1 21 0.0996 27 12.02 27 10.76 2 0.16 2 0.15 27 1.56 P 2 22 0.1100 13 3.49 12 3.07 15 0.39 15 0.37 23 0.76 C 2 23 0.1124 6 0.76 6 0.69 17 0.41 17 0.39 8 0.18 C 1 24 0.1132 1 0.25 1 0.24 17 0.41 17 0.39 7 0.17 P 1 25 0.1133 8 1.38 8 1.24 15 0.39 15 0.37 22 0.76 H 1 26 0.1230 14 3.58 14 3.56 8 0.34 6 0.30 14 0.33 H 2 27 0.1249 17 4.13 17 4.05 8 0.34 6 0.30 13 0.33 T 1 28 0.1929 16 3.82 16 3.73 6 0.33 8 0.31 1 0.01 M 2 29 0.2526 26 11.31 21 9.39 25 1.14 27 1.09 3 0.14 M 1 30 0.2615 10 1.75 10 1.78 25 1.14 27 1.09 3 0.14 D 2 1 0.0190 2 29.26 2 28.64 5 1.57 5 1.43 3 0.61 I 2 2 0.0214 3 41.60 3 40.35 6 2.19 6 2.15 5 1.01 V 2 3 0.0222 6 64.47 6 61.24 2 0.34 2 0.32 1 0.39 S 2 4 0.0503 4 43.56 4 46.05 4 0.89 4 0.88 4 0.71 G 2 5 0.0909 5 57.88 5 59.55 3 0.57 3 0.55 2 0.57 U 2 6 0.1007 1 22.39 1 20.10 1 0.16 1 0.15 6 1.30 Table 82: Tabulation of ranked results for Class A — Right Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 100 000, as seen in Table 17. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 73. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 61. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Identification Mean Median FNIR 181 Search Enrollment Mean Median RAM Letter Sub. # V 1 1 0.0034 5 9.50 5 9.29 4 0.68 3 0.64 5 11.77 I 1 2 0.0058 9 18.92 8 17.87 9 2.96 9 2.93 6 15.83 J 1 3 0.0143 6 14.00 6 13.64 3 0.66 4 0.65 3 8.30 L 1 4 0.0146 1 2.20 2 2.19 1 0.11 1 0.11 8 18.76 O 1 5 0.0229 7 15.34 7 14.46 2 0.55 2 0.53 4 9.05 K 1 6 0.0360 8 18.32 9 18.01 8 2.11 8 2.09 9 61.81 C 1 7 0.0368 2 2.39 1 2.08 5 0.80 5 0.78 2 4.87 C 2 8 0.0374 4 6.34 4 6.35 5 0.80 5 0.78 1 4.87 G 1 9 0.0515 3 6.27 3 5.22 7 1.13 7 1.09 7 16.38 Q 1 1 0.0027 21 213.08 21 212.69 16 1.13 15 1.08 10 9.14 Q 2 1 0.0027 19 163.65 19 161.02 16 1.13 15 1.08 9 9.14 V 2 3 0.0028 18 133.45 18 127.65 10 0.68 9 0.64 13 11.77 I 2 4 0.0030 25 385.14 25 338.88 27 4.37 27 4.30 21 30.83 D 2 4 0.0030 23 234.52 23 237.43 26 3.12 26 2.84 17 18.58 D 1 4 0.0030 16 73.01 15 70.99 25 3.10 25 2.84 17 18.58 L 2 7 0.0072 3 23.54 3 23.42 3 0.33 3 0.31 26 53.98 J 2 8 0.0143 7 36.19 7 33.35 9 0.66 10 0.65 7 8.30 S 2 9 0.0195 26 429.02 26 495.50 18 1.77 18 1.75 14 14.82 E 2 10 0.0202 27 500.30 27 518.11 11 0.70 11 0.70 23 33.41 E 1 11 0.0207 2 22.27 1 16.35 11 0.70 11 0.70 22 33.40 O 2 12 0.0214 10 45.52 10 43.53 4 0.55 4 0.53 8 9.05 0.0281 1 14.82 S 1 13 20.11 2 23.00 18 1.77 18 1.75 14 K 2 14 0.0286 6 32.68 6 32.93 20 2.11 20 2.09 27 61.81 G 2 15 0.0311 22 227.27 22 221.16 15 1.13 17 1.09 16 16.38 P 2 16 0.0333 17 114.37 17 101.16 13 0.77 13 0.72 19 21.41 U 1 17 0.0336 11 47.60 11 45.51 1 0.30 1 0.30 25 44.63 U 2 18 0.0358 24 252.04 24 240.40 1 0.30 1 0.30 24 37.35 T 2 19 0.0366 9 38.91 9 37.23 5 0.65 7 0.62 1 0.01 P 1 20 0.0370 14 63.41 14 63.65 13 0.77 13 0.72 20 21.42 F 1 21 0.0386 13 59.30 13 60.66 21 2.23 21 2.15 5 4.86 F 2 22 0.0412 15 71.45 16 73.95 21 2.23 21 2.15 5 4.86 H 2 23 0.0684 12 53.77 12 52.25 7 0.66 5 0.60 11 9.36 H 1 24 0.0686 8 37.65 8 36.71 7 0.66 5 0.60 12 9.36 M 2 25 NA 20 183.20 20 171.24 23 2.25 23 2.16 4 3.76 M 1 26 NA 5 32.59 5 31.44 23 2.25 23 2.16 3 3.75 T 1 27 NA 4 26.96 4 25.68 5 0.65 7 0.62 1 0.01 Table 83: Tabulation of ranked results for Class A — Left and Right Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 1 600 000, as seen in Table 18. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 74. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 62. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 182 FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR Search Enrollment Mean Median RAM I 2 1 0.0094 18 51.59 13 38.26 30 7.68 30 7.67 21 108.71 Q 1 2 0.0098 17 51.33 19 53.24 20 3.41 20 3.41 1 7.54 Q 2 3 0.0099 16 51.05 20 54.02 20 3.41 20 3.41 2 7.54 I 1 4 0.0116 23 63.06 21 55.96 25 6.50 25 6.50 5 49.68 D 2 5 0.0142 21 59.65 22 58.10 24 4.18 24 4.17 12 79.43 D 1 6 0.0163 20 53.09 18 52.23 14 2.16 14 2.15 13 79.43 E 2 7 0.0187 12 41.57 10 34.37 15 2.21 15 2.21 28 318.03 V 2 8 0.0190 14 48.01 15 47.97 3 1.21 7 1.22 9 63.53 V 1 9 0.0192 10 36.28 11 36.07 3 1.21 7 1.22 8 63.53 J 2 10 0.0236 27 80.54 27 74.38 7 1.22 3 1.21 19 101.00 O 2 11 0.0254 24 78.80 26 73.15 5 1.22 5 1.21 18 101.00 O 1 12 0.0257 11 38.17 12 37.01 5 1.22 5 1.21 17 101.00 E 1 13 0.0259 1 3.71 1 2.82 11 1.95 11 1.95 14 79.72 L 2 14 0.0276 5 12.52 5 12.37 1 0.33 1 0.33 27 177.49 J 1 15 0.0287 7 26.88 7 25.61 7 1.22 3 1.21 20 101.00 L 1 16 0.0288 3 5.51 3 5.56 2 1.13 2 1.11 22 119.79 G 2 17 0.0325 22 62.98 23 59.49 18 2.81 18 2.83 25 156.12 G 1 18 0.0371 6 19.65 6 16.24 18 2.81 18 2.83 26 156.12 U 2 19 0.0461 29 90.50 30 89.07 16 2.45 16 2.41 30 540.60 U 1 20 0.0500 28 81.85 28 80.03 17 2.46 17 2.42 29 440.67 C 2 21 0.0647 4 6.62 4 6.74 12 2.13 12 2.12 4 46.49 C 1 22 0.0654 2 3.71 2 3.74 12 2.13 12 2.12 3 46.49 H 1 23 0.0998 15 49.13 16 49.75 9 1.68 9 1.67 16 86.46 H 2 24 0.1008 19 52.05 17 51.77 9 1.68 9 1.67 15 86.46 S 1 25 0.1089 25 78.95 24 71.69 22 3.97 22 3.96 24 150.35 S 2 26 0.1133 26 79.01 25 71.77 22 3.97 22 3.96 23 150.35 M 2 27 0.1634 30 90.51 29 87.21 26 6.94 28 6.73 7 57.53 F 2 28 0.1681 13 46.18 14 43.59 28 6.94 26 6.73 10 77.02 F 1 29 0.1684 8 31.92 8 30.09 28 6.94 26 6.73 11 77.02 M 1 30 0.1736 9 32.39 9 31.17 26 6.94 28 6.73 6 57.53 Table 84: Tabulation of ranked results for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 75. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR 183 Search Enrollment Mean Median RAM I 2 1 0.0045 16 49.18 14 41.13 30 7.65 30 7.65 21 108.71 D 2 2 0.0052 20 58.67 21 57.19 24 4.15 24 4.13 12 79.43 Q 1 3 0.0057 22 62.95 23 64.04 20 3.41 20 3.41 1 7.54 Q 2 3 0.0057 21 61.29 22 61.94 20 3.41 20 3.41 2 7.54 I 1 5 0.0058 23 63.57 20 55.87 25 6.47 25 6.47 5 49.68 D 1 6 0.0072 19 53.78 19 53.32 14 2.15 14 2.14 13 79.43 E 2 7 0.0083 12 40.27 10 34.47 15 2.18 15 2.18 28 318.00 V 1 8 0.0106 10 35.66 11 35.43 3 1.19 3 1.19 8 63.53 V 2 9 0.0110 15 48.47 16 48.51 3 1.19 3 1.19 9 63.53 J 2 10 0.0126 27 77.67 27 74.09 5 1.21 5 1.20 19 101.00 O 2 11 0.0132 26 77.37 26 73.93 7 1.21 7 1.20 20 101.00 O 1 12 0.0142 11 38.08 12 35.61 7 1.21 7 1.20 17 101.00 E 1 13 0.0151 2 4.68 1 2.97 11 1.92 11 1.91 14 79.72 J 1 14 0.0156 7 25.86 7 24.14 5 1.21 5 1.20 18 101.00 L 1 15 0.0167 3 5.49 3 5.47 2 1.12 2 1.11 22 119.78 G 2 16 0.0198 13 41.12 13 37.28 18 2.76 18 2.78 25 156.12 L 2 17 0.0202 6 12.66 6 12.66 1 0.33 1 0.33 27 177.49 G 1 18 0.0212 5 11.80 5 10.13 18 2.76 18 2.78 26 156.12 U 1 19 0.0266 29 89.71 30 89.26 16 2.43 16 2.39 29 440.67 U 2 20 0.0273 28 88.88 28 80.11 17 2.43 17 2.39 30 540.59 S 1 21 0.0369 25 74.29 24 70.54 22 3.89 22 3.88 23 150.35 S 2 22 0.0381 24 73.96 25 70.84 22 3.89 22 3.88 24 150.35 C 2 23 0.0392 4 6.54 4 6.50 12 2.10 12 2.09 3 46.49 C 1 24 0.0403 1 3.70 2 3.67 12 2.10 12 2.09 4 46.49 H 1 25 0.0641 17 50.97 17 50.16 9 1.64 9 1.63 16 86.46 H 2 26 0.0647 18 52.86 18 52.84 9 1.64 9 1.63 15 86.46 M 2 27 0.1155 30 91.42 29 87.41 28 6.88 28 6.69 6 57.53 F 2 28 0.1220 14 46.58 15 45.92 26 6.86 26 6.68 10 77.02 F 1 29 0.1222 8 32.28 8 31.87 26 6.86 26 6.68 11 77.02 M 1 30 0.1259 9 33.43 9 32.03 28 6.88 28 6.69 7 57.53 Table 85: Tabulation of ranked results for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 76. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 184 FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR Search Enrollment Mean Median RAM I 2 1 0.0015 19 57.94 17 46.13 30 15.30 30 15.33 21 108.71 Q 1 2 0.0021 20 63.99 22 65.02 20 6.84 20 6.84 2 7.54 Q 2 3 0.0022 18 53.88 19 53.40 20 6.84 20 6.84 1 7.54 I 1 3 0.0022 15 47.35 13 37.19 25 12.94 25 12.96 5 49.68 D 2 5 0.0024 24 74.21 21 63.05 24 8.31 24 8.28 13 79.43 D 1 6 0.0031 21 69.35 20 61.24 14 4.29 14 4.28 12 79.43 V 1 7 0.0036 11 39.24 15 38.93 3 2.40 7 2.40 9 63.53 V 2 7 0.0036 17 52.97 18 52.80 3 2.40 7 2.40 8 63.53 J 2 9 0.0047 26 75.92 24 69.28 7 2.42 5 2.40 17 101.00 E 2 10 0.0049 12 39.56 10 33.09 15 4.39 15 4.38 28 317.99 O 2 11 0.0051 25 74.57 23 68.46 5 2.42 3 2.40 19 101.00 L 1 12 0.0054 4 10.37 4 10.48 2 2.24 2 2.23 22 119.77 O 1 13 0.0057 13 40.42 14 38.34 5 2.42 3 2.40 20 101.00 L 2 14 0.0062 6 20.85 6 20.78 1 0.66 1 0.66 27 177.49 E 1 15 0.0063 3 8.18 2 6.12 11 3.86 11 3.85 14 79.72 J 1 16 0.0068 7 27.52 7 26.15 7 2.42 5 2.40 18 101.00 G 2 17 0.0084 14 42.86 11 34.55 18 5.59 18 5.62 25 156.12 G 1 18 0.0106 1 5.52 1 3.89 18 5.59 18 5.62 26 156.12 U 2 19 0.0124 30 91.04 28 82.40 16 4.89 17 4.83 30 540.59 U 1 20 0.0139 29 86.88 30 83.62 17 4.91 16 4.82 29 440.67 S 1 21 0.0160 28 85.05 29 83.33 22 7.82 22 7.83 23 150.35 S 2 22 0.0190 27 84.02 27 82.13 22 7.82 22 7.83 24 150.35 H 1 23 0.0349 22 70.12 25 70.38 9 3.31 9 3.31 16 86.46 H 2 24 0.0361 23 72.35 26 73.06 9 3.31 9 3.31 15 86.46 M 2 25 0.0882 16 49.50 16 45.91 28 13.79 28 13.53 6 57.53 F 2 26 0.0901 10 38.37 12 36.36 26 13.78 26 13.49 11 77.02 M 1 27 0.0904 9 28.89 8 27.15 28 13.79 28 13.53 7 57.53 F 1 28 0.0910 8 28.72 9 27.27 26 13.78 26 13.49 10 77.02 C 2 29 NA 5 10.63 5 10.70 12 4.22 12 4.20 4 46.49 C 1 30 NA 2 7.36 3 6.40 12 4.22 12 4.20 3 46.49 Table 86: Tabulation of ranked results for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 77. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR 185 Search Enrollment Mean Median RAM I 2 1 0.0009 20 60.01 16 43.86 30 19.36 30 19.36 21 108.71 Q 1 2 0.0012 14 48.85 14 42.92 20 8.89 20 8.92 1 7.54 Q 2 2 0.0012 24 71.67 24 66.01 20 8.89 20 8.92 2 7.54 I 1 2 0.0012 6 24.57 7 19.07 25 16.50 25 16.52 5 49.68 D 2 2 0.0012 19 54.37 18 46.70 24 11.90 24 11.86 13 79.43 D 1 6 0.0020 17 52.42 17 45.15 17 6.19 17 6.16 12 79.43 V 2 7 0.0024 15 49.72 19 49.30 3 3.35 7 3.36 9 63.53 E 2 7 0.0024 18 52.88 15 43.61 16 5.95 16 5.93 28 317.95 V 1 9 0.0027 10 35.51 10 34.96 3 3.35 7 3.36 8 63.53 L 1 10 0.0031 5 14.42 5 14.50 2 3.05 2 3.05 22 119.79 L 2 11 0.0033 8 28.56 9 28.59 1 0.88 1 0.88 27 177.48 J 2 11 0.0033 22 64.38 22 60.13 7 3.38 3 3.35 17 101.00 O 2 13 0.0035 21 63.87 21 60.02 5 3.38 5 3.36 20 101.00 G 2 14 0.0040 9 31.67 6 16.26 18 7.89 18 7.90 26 156.12 O 1 15 0.0041 11 37.04 11 35.64 5 3.38 5 3.36 19 101.00 E 1 16 0.0043 2 8.76 2 6.76 13 5.35 13 5.34 14 79.73 J 1 17 0.0049 7 26.31 8 25.38 7 3.38 3 3.35 18 101.00 G 1 18 0.0062 1 6.33 1 4.26 18 7.89 18 7.90 25 156.12 U 1 19 0.0099 30 88.83 28 86.60 15 5.90 15 5.82 29 440.68 S 1 20 0.0108 28 86.50 29 87.60 22 10.24 22 10.31 24 150.35 S 2 21 0.0136 29 88.70 30 88.46 22 10.24 22 10.31 23 150.35 U 2 22 0.0141 25 80.14 25 75.28 14 5.80 14 5.73 30 540.60 H 1 23 0.0203 26 82.43 26 82.66 9 4.37 9 4.37 15 86.46 H 2 24 0.0204 27 85.74 27 86.56 9 4.37 9 4.37 16 86.46 M 2 25 0.0515 23 70.45 23 65.91 26 18.36 28 18.11 6 57.53 M 1 26 0.0543 13 41.37 13 38.87 26 18.36 28 18.11 6 57.53 F 1 27 0.0591 12 39.12 12 37.56 28 18.36 26 18.07 10 77.02 F 2 27 0.0591 16 51.59 20 49.34 28 18.36 26 18.07 10 77.02 C 2 29 NA 4 10.28 4 10.21 11 5.16 11 5.13 4 46.49 C 1 30 NA 3 9.00 3 7.92 11 5.16 11 5.13 3 46.49 Table 87: Tabulation of ranked results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 78. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 186 FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR Search Enrollment Mean Median RAM I 2 1 0.0010 15 56.93 10 38.28 30 17.04 30 16.99 11 110.22 Q 1 2 0.0011 10 44.75 9 36.86 20 9.14 20 9.11 2 13.33 D 2 2 0.0011 25 84.92 26 74.36 24 13.69 24 13.66 15 132.37 Q 2 4 0.0013 9 43.67 8 35.35 20 9.14 20 9.11 1 13.33 I 1 4 0.0013 23 79.70 16 52.41 29 16.71 29 16.67 4 84.42 D 1 6 0.0015 28 87.07 25 73.93 17 7.12 17 7.10 16 132.37 V 1 7 0.0024 13 47.76 14 47.02 7 3.69 7 3.66 13 121.80 V 2 7 0.0024 17 63.19 19 62.26 7 3.69 7 3.66 12 121.79 O 1 9 0.0025 21 73.31 23 68.54 4 3.63 3 3.61 19 181.00 O 2 10 0.0027 24 82.13 24 73.41 4 3.63 3 3.61 22 181.00 J 2 10 0.0027 26 85.17 28 77.75 6 3.64 6 3.62 21 181.00 J 1 12 0.0047 12 46.32 12 43.57 3 3.62 5 3.62 20 181.00 E 2 13 0.0048 27 85.50 17 52.91 12 5.21 12 5.20 28 573.25 E 1 14 0.0088 2 15.94 2 12.77 11 5.20 11 5.20 14 127.39 L 2 15 0.0095 6 21.26 6 23.53 1 3.51 1 3.48 27 274.95 L 1 16 0.0102 3 17.53 3 17.51 1 3.51 1 3.48 23 192.19 U 2 17 0.0155 20 67.98 22 68.30 15 6.13 15 5.91 29 811.45 U 1 18 0.0163 22 75.14 27 74.94 15 6.13 15 5.91 29 811.45 H 2 19 0.0275 18 64.90 20 65.36 9 5.16 9 5.16 18 144.02 G 2 20 0.0276 7 37.15 7 23.69 18 8.35 18 8.32 26 274.35 H 1 20 0.0276 19 67.91 21 68.07 9 5.16 9 5.16 17 144.02 S 1 22 0.0311 29 87.76 29 86.56 22 10.91 22 10.87 24 260.37 G 1 23 0.0368 1 11.21 1 7.71 18 8.35 18 8.32 25 274.35 C 2 24 0.0711 4 18.31 5 18.38 13 5.84 13 5.80 10 92.86 F 1 25 0.0734 14 50.99 15 51.69 27 15.96 27 15.87 7 90.93 F 2 25 0.0734 16 60.34 18 60.86 27 15.96 27 15.87 6 90.93 M 2 27 0.0826 11 45.96 13 46.70 25 15.91 25 15.83 8 90.93 M 1 28 0.0934 8 42.60 11 43.37 25 15.91 25 15.83 5 90.93 S 2 29 0.1680 30 89.24 30 91.74 22 10.91 22 10.87 3 61.04 C 1 30 NA 5 21.04 4 18.30 13 5.84 13 5.80 9 92.86 Table 88: Tabulation of ranked results for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 5 000 000, as seen in Table 20. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 79. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 64. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Letter Sub. # Identification Mean Median FNIR 187 Search Enrollment Mean Median RAM I 1 1 0.0013 24 79.04 15 54.56 25 20.21 25 20.23 4 113.80 Q 2 2 0.0014 27 83.35 25 74.22 17 12.95 17 13.06 2 20.22 I 2 2 0.0014 9 40.53 7 30.82 24 18.72 24 18.73 11 137.03 D 1 4 0.0015 17 65.97 17 58.92 23 17.44 21 17.39 10 132.40 Q 1 5 0.0017 26 83.25 26 74.40 17 12.95 17 13.06 1 20.22 V 1 5 0.0017 10 40.94 10 40.74 9 8.48 9 8.50 17 234.03 D 2 7 0.0018 30 86.39 27 74.97 30 30.43 30 30.47 9 132.37 V 2 8 0.0019 16 65.47 18 65.07 9 8.48 9 8.50 18 234.03 O 2 9 0.0033 19 72.55 21 69.30 6 6.80 6 6.71 23 303.13 J 2 9 0.0033 20 72.62 19 67.98 8 6.82 8 6.74 24 303.13 O 1 11 0.0034 15 59.01 14 54.42 6 6.80 6 6.71 25 303.13 E 2 12 0.0050 14 57.02 11 42.73 11 9.89 11 9.91 30 930.48 J 1 13 0.0051 8 36.09 9 33.02 5 6.78 5 6.69 22 303.13 L 2 14 0.0083 3 19.52 5 20.25 3 4.36 3 4.36 26 367.82 C 2 15 0.0085 5 25.91 6 21.97 13 10.79 13 10.69 15 183.36 C 1 16 0.0094 4 25.25 4 20.15 13 10.79 13 10.69 14 183.36 L 1 17 0.0097 7 31.42 8 31.68 3 4.36 3 4.36 21 280.49 E 1 18 0.0106 1 9.52 2 8.63 12 9.90 12 9.92 16 191.66 H 2 19 0.0199 29 84.26 29 84.50 15 12.08 15 12.17 12 144.02 H 1 20 0.0201 28 84.14 30 84.51 15 12.08 15 12.17 13 144.02 G 2 21 0.0333 6 30.61 3 19.46 19 15.55 19 15.50 19 261.29 U 2 22 0.0351 23 74.39 24 72.48 1 2.94 1 2.87 28 806.17 U 1 23 0.0358 25 82.76 28 82.61 1 2.94 1 2.87 28 806.17 G 1 24 0.0447 2 11.67 1 7.78 19 15.55 19 15.50 20 261.29 F 1 25 0.0536 13 55.63 16 55.79 28 21.07 28 20.92 7 130.29 F 2 25 0.0536 18 68.61 20 68.20 28 21.07 28 20.92 6 130.29 M 2 27 0.0716 12 48.71 13 48.80 26 21.02 26 20.89 5 130.29 M 1 28 0.0783 11 47.38 12 48.08 26 21.02 26 20.89 8 130.29 S 1 29 0.0860 22 74.01 22 69.71 21 17.43 22 17.57 27 382.88 S 2 30 0.2462 21 72.95 23 70.77 21 17.43 22 17.57 3 78.28 Table 89: Tabulation of ranked results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 5 000 000, as seen in Table 20. The Search Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 80. Identification and Search Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 65. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on the FNIR column-wise ranking. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 188 J FPVTE – F INGERPRINT M ATCHING Relative Combined Results The tables in Appendix I show the detailed values for FNIR at a fixed FPIR of 10−3 , identification time, search template enrollment time, and RAM consumption. While the column-wise ranking for each variable is given, it can be difficult to visualize relative comparisons among submissions for each of these values. Use of star plots can help with this visualization. For more information on how to read star plots, please refer to the explanation in Appendix K. Class A plots are in Figures 99 through 101, Class B plots are in Figures 102 through 105, and Class C plots are in Figures 106 through 107. Some notable observations based on the plotted shape include that: . For almost all finger combinations, I uses the longest enrollment times while providing the highest accuracy. . D’s submission strike a balance between identification time and enrollment time in order to provide high accuracy. . Class A results are nearly identical, regardless of which index finger combinations are used. . J1, L1, Q, O1, and V1 provide high accuracy while limiting time and computational resources for most finger combinations. . Shapes formed by participants appear to remain similar regardless of finger combination. . Tradeoffs between a participant’s two submissions are very easily seen. For instance, in class Class A, V1 and S1 use significantly shorter identification times than V2 and S2 respectively, but result in similar relative accuracies. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 189 E2 FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● Identification Time ● ● ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time ● ● ● ● ● K2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time ● ● Enrollment Time RAM Consumption K1 RAM Consumption Enrollment Time J2 RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption Enrollment Time ● Identification Time ● ● Identification Time ● ● ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR Identification Time RAM Consumption FPVTE – F INGERPRINT M ATCHING ● ● ● ● O1 O2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time S2 FNIR ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time Identification Time Enrollment Time ● Identification Time ● ● Identification Time ● Identification Time ● ● Identification Time ● RAM Consumption S1 FNIR RAM Consumption Q2 FNIR RAM Consumption Q1 FNIR RAM Consumption P2 FNIR RAM Consumption P1 FNIR Identification Time RAM Consumption Enrollment Time ● ● Identification Time ● ● Identification Time Enrollment Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Enrollment Time RAM Consumption M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time L2 RAM Consumption Enrollment Time L1 RAM Consumption Enrollment Time Identification Time RAM Consumption ● Enrollment Time ● ● ● ● Enrollment Time T1 T2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption RAM Consumption ● ● ● ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time Enrollment Time ● ● Identification Time ● ● ● Identification Time ● Identification Time RAM Consumption ● Enrollment Time Figure 99: Star plot of combined results for Class A — Left Index. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 81, with the exception of FNIR, whose log10 values were scaled. Any values printed as NA were set to 1 before scaling. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING E2 FNIR ● ● Enrollment Time ● ● ● ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time Enrollment Time H2 FNIR ● ● ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● ● ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption H1 FNIR RAM Consumption G2 FNIR RAM Consumption G1 FNIR RAM Consumption F2 FNIR RAM Consumption F1 FNIR Identification Time RAM Consumption Enrollment Time ● ● ● ● ● ● ● K2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time ● ● Enrollment Time RAM Consumption K1 RAM Consumption Enrollment Time J2 RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption Enrollment Time ● Identification Time ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR Identification Time RAM Consumption 190 ● ● ● ● O2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● S1 S2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time Identification Time Enrollment Time ● ● Identification Time ● ● Identification Time Enrollment Time ● ● Identification Time ● ● Identification Time ● ● Enrollment Time RAM Consumption Enrollment Time Q2 RAM Consumption Enrollment Time Q1 RAM Consumption Enrollment Time P2 RAM Consumption Enrollment Time P1 RAM Consumption Enrollment Time ● Identification Time ● ● Identification Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Enrollment Time RAM Consumption O1 RAM Consumption M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time L2 RAM Consumption Enrollment Time L1 Identification Time RAM Consumption Enrollment Time Identification Time RAM Consumption ● Enrollment Time ● ● ● ● Enrollment Time T1 T2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption RAM Consumption ● ● ● ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time Enrollment Time ● ● Identification Time ● ● ● Identification Time ● Identification Time RAM Consumption ● Enrollment Time Figure 100: Star plot of combined results for Class A — Right Index. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 82, with the exception of FNIR, whose log10 values were scaled. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 191 E2 FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● K2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption K1 RAM Consumption Enrollment Time J2 RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption Enrollment Time ● Identification Time ● ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time Identification Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR Identification Time RAM Consumption FPVTE – F INGERPRINT M ATCHING ● ● ● ● O1 O2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● Enrollment Time Enrollment Time ● ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time S2 FNIR ● ● ● ● ● Enrollment Time Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● Identification Time ● RAM Consumption S1 FNIR RAM Consumption Q2 FNIR RAM Consumption Q1 FNIR RAM Consumption P2 FNIR RAM Consumption P1 FNIR Identification Time RAM Consumption ● Identification Time ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● Enrollment Time RAM Consumption M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time L2 RAM Consumption Enrollment Time L1 RAM Consumption Enrollment Time Identification Time RAM Consumption ● Enrollment Time Enrollment Time T1 T2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption RAM Consumption ● ● ● ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time Enrollment Time ● ● Identification Time ● ● ● Identification Time ● Identification Time RAM Consumption ● Enrollment Time Figure 101: Star plot of combined results for Class A — Left and Right Index. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 21, with the exception of FNIR, whose log10 values were scaled. Any values printed as NA were set to 1 before scaling. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 192 FPVTE – F INGERPRINT M ATCHING ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● J2 L1 L2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● RAM Consumption ● Identification Time ● ● Enrollment Time Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● ● RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption ● Enrollment Time RAM Consumption ● Enrollment Time Identification Time ● Identification Time ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption E2 FNIR RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR ● ● ● ● ● ● Q1 Q2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Enrollment Time Enrollment Time Enrollment Time S2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● Enrollment Time ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● ● ● ● ● Identification Time ● ● Enrollment Time Identification Time Enrollment Time ● ● Identification Time ● ● Identification Time ● Identification Time ● ● RAM Consumption Enrollment Time S1 RAM Consumption Enrollment Time RAM Consumption ● RAM Consumption ● Identification Time RAM Consumption ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time ● Identification Time Identification Time ● Enrollment Time RAM Consumption O2 RAM Consumption Enrollment Time O1 RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time Enrollment Time Figure 102: Star plot of combined results for Class B — Left Slap. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 84, with the exception of FNIR, whose log10 values were scaled. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 193 ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● J2 L1 L2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● RAM Consumption ● ● ● Identification Time ● ● Enrollment Time Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● ● RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption ● Enrollment Time RAM Consumption ● Enrollment Time ● Identification Time ● Identification Time ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption E2 FNIR RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR ● ● ● ● ● ● Q1 Q2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Enrollment Time Enrollment Time Enrollment Time S2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● ● ● ● ● ● Identification Time Enrollment Time ● Enrollment Time Identification Time ● ● ● Identification Time ● ● Identification Time ● ● Identification Time ● RAM Consumption Enrollment Time S1 RAM Consumption Enrollment Time RAM Consumption ● RAM Consumption ● Identification Time RAM Consumption ● Identification Time ● Identification Time ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption O2 RAM Consumption Enrollment Time O1 RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time Enrollment Time Figure 103: Star plot of combined results for Class B — Right Slap. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 85, with the exception of FNIR, whose log10 values were scaled. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 194 FPVTE – F INGERPRINT M ATCHING ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● Enrollment Time Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time L2 FNIR ● ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time ● ● RAM Consumption L1 FNIR RAM Consumption J2 FNIR RAM Consumption J1 FNIR RAM Consumption I2 FNIR RAM Consumption I1 FNIR Identification Time RAM Consumption ● Enrollment Time ● ● ● Identification Time ● Enrollment Time ● ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● RAM Consumption E2 FNIR RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR ● RAM Consumption C2 FNIR ● RAM Consumption C1 FNIR ● ● ● ● ● ● Q1 Q2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time Enrollment Time ● ● ● ● ● Enrollment Time Enrollment Time ● ● ● ● Enrollment Time Enrollment Time ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time Identification Time Enrollment Time ● Identification Time ● ● Identification Time ● Identification Time ● RAM Consumption V2 FNIR RAM Consumption V1 FNIR RAM Consumption U2 FNIR RAM Consumption U1 FNIR RAM Consumption S2 FNIR ● ● ● S1 ● ● ● ● FNIR Identification Time RAM Consumption ● ● Identification Time ● ● ● Identification Time ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● Enrollment Time RAM Consumption O2 RAM Consumption Enrollment Time O1 RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time ● ● ● Enrollment Time Figure 104: Star plot of combined results for Class B — Left and Right Slap. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 86, with the exception of FNIR, whose log10 values were scaled. Any values printed as NA were set to 1 before scaling. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING 195 ● ● ● ● ● ● ● ● ● ● ● ● ● ● H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● Enrollment Time Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time L2 FNIR ● ● ● ● ● ● ● ● ● ● Identification Time ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time ● ● RAM Consumption L1 FNIR RAM Consumption J2 FNIR RAM Consumption J1 FNIR RAM Consumption I2 FNIR RAM Consumption I1 FNIR Identification Time RAM Consumption ● Enrollment Time ● ● ● Identification Time ● Enrollment Time ● ● Identification Time ● ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption H1 RAM Consumption Enrollment Time G2 RAM Consumption Enrollment Time G1 RAM Consumption Enrollment Time F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● RAM Consumption E2 FNIR RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR ● RAM Consumption C2 FNIR ● RAM Consumption C1 FNIR ● ● ● ● ● ● Q1 Q2 FNIR FNIR FNIR FNIR FNIR FNIR ● Enrollment Time Enrollment Time ● ● ● ● ● ● Enrollment Time Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Enrollment Time V2 FNIR ● Enrollment Time ● ● ● Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time ● ● ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time ● ● Identification Time ● ● Identification Time ● RAM Consumption V1 FNIR RAM Consumption U2 FNIR RAM Consumption U1 FNIR RAM Consumption S2 FNIR RAM Consumption S1 FNIR Identification Time RAM Consumption ● ● ● Identification Time ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● Enrollment Time RAM Consumption O2 RAM Consumption Enrollment Time O1 RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time Enrollment Time Figure 105: Star plot of combined results for Class B — Identification Flats. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 22, with the exception of FNIR, whose log10 values were scaled. Any values printed as NA were set to 1 before scaling. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 196 FPVTE – F INGERPRINT M ATCHING C1 C2 D1 D2 E1 E2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● RAM Consumption ● ● RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time RAM Consumption ● ● ● ● ● ● G1 G2 H1 H2 FNIR FNIR FNIR FNIR FNIR FNIR ● Enrollment Time ● ● ● ● ● Enrollment Time Enrollment Time ● ● ● Enrollment Time ● ● ● ● Enrollment Time L2 FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● Enrollment Time O1 O2 Q1 Q2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time Enrollment Time ● ● ● ● Enrollment Time Enrollment Time ● ● ● ● RAM Consumption RAM Consumption ● ● ● Enrollment Time Identification Time ● ● ● Enrollment Time Identification Time ● Identification Time ● Identification Time ● Identification Time Identification Time ● ● RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption ● Enrollment Time RAM Consumption ● Enrollment Time ● Identification Time ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time ● ● RAM Consumption L1 FNIR RAM Consumption J2 FNIR RAM Consumption J1 FNIR RAM Consumption I2 FNIR RAM Consumption I1 FNIR Identification Time RAM Consumption ● Enrollment Time ● ● Identification Time ● ● Identification Time ● ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● Enrollment Time RAM Consumption F2 RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time RAM Consumption Enrollment Time RAM Consumption Enrollment Time ● ● ● ● Enrollment Time S1 S2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Enrollment Time Enrollment Time ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● ● RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption ● ● ● ● ● Identification Time ● ● Identification Time Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● Identification Time RAM Consumption ● ● Enrollment Time Figure 106: Star plot of combined results for Class C — Ten-Finger Plain-to-Plain. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 88, with the exception of FNIR, whose log10 values were scaled. Any values printed as NA were set to 1 before scaling. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 197 E2 FNIR ● ● ● ● ● ● ● ● ● Identification Time ● ● ● Identification Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● RAM Consumption E1 FNIR RAM Consumption D2 FNIR RAM Consumption D1 FNIR RAM Consumption C2 FNIR RAM Consumption C1 FNIR Identification Time RAM Consumption FPVTE – F INGERPRINT M ATCHING ● ● ● ● ● F2 G1 G2 H1 H2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● RAM Consumption ● ● ● ● ● ● ● ● ● ● ● ● ● L2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● Q2 FNIR FNIR FNIR FNIR FNIR FNIR ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Enrollment Time Enrollment Time Enrollment Time Enrollment Time Enrollment Time S1 S2 U1 U2 V1 V2 FNIR FNIR FNIR FNIR FNIR FNIR Identification Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time Identification Time ● Enrollment Time RAM Consumption Q1 RAM Consumption Enrollment Time O2 RAM Consumption Enrollment Time O1 RAM Consumption Enrollment Time M2 RAM Consumption Enrollment Time M1 RAM Consumption Enrollment Time ● Identification Time ● ● ● Identification Time ● Identification Time ● ● ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption L1 RAM Consumption Enrollment Time J2 RAM Consumption Enrollment Time J1 RAM Consumption Enrollment Time I2 RAM Consumption Enrollment Time I1 RAM Consumption Enrollment Time Identification Time ● ● ● Identification Time ● Identification Time ● ● Identification Time ● Identification Time Identification Time ● ● Enrollment Time RAM Consumption Enrollment Time F1 RAM Consumption Enrollment Time RAM Consumption Enrollment Time RAM Consumption Enrollment Time RAM Consumption Enrollment Time Enrollment Time ● Enrollment Time Enrollment Time ● ● Enrollment Time ● ● ● ● Enrollment Time RAM Consumption ● RAM Consumption RAM Consumption RAM Consumption RAM Consumption ● ● ● ● ● ● Identification Time ● ● Identification Time Enrollment Time ● ● Identification Time ● ● ● Identification Time ● Identification Time ● Identification Time RAM Consumption ● Enrollment Time Figure 107: Star plot of combined results for Class C — Ten-Finger Rolled-to-Rolled. The values in this plot have been independently scaled from 0 to 1 from the values printed in Table 23, with the exception of FNIR, whose log10 values were scaled. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 198 K FPVTE – F INGERPRINT M ATCHING Relative Accuracy and Number of Fingers In Section 7, DETs, tables, and other data visualization methods were used to show exact FNIR values at FPIR = 10−3 . While this data is important for exact comparisons, it is difficult to quickly compare relative accuracy among submissions, or to see how a submission fares across finger combinations. To facilitate these comparisons, star plots [1] (also called spider or radar plots) are included in this section. These plots are used to quickly examine relative values among multiple variables. In each star plot, values are plotted along multiple radii, where each radius represents a single variable. The plot area is delineated with three circles. The points on the circumference of these circles indicate different values where they intersect the radii. The intersection with the smallest circle represents 0, the next-largest circle (dashed-blue) represents 0.5, and the largest circle (dashed-gray) represents 1. All values plotted have been scaled between 0 and 1 against other values for the same variable in order to fit within the largest circle. These plotted values were then connected to adjacent plotted values, creating a polygon. Many traits may be quickly inferred by the shape of these polygons: . Regular polygons indicate submissions that have similar accuracy for all plotted variables. . A submission that is the most accurate in terms of every variable would inscribe the center circle and the least accurate would inscribe the largest circle. . The smaller the polygon, the more accurate the submission. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING K.1 199 Class A Relative FNIR results for left index, right index, and left and right index searches are shown in Figure 108. Note that the enrollment set sizes differed between single index and two-index searches. Some notable observations include: . The most accurate submissions (D, I, Q, V) are instantly recognizable by their small shape. . Most submissions have near-identical accuracy, regardless of being provided one or two index fingers, as indicated by the equilateral triangle drawn. This may have impacts on data collection and storage. Extreme examples include D, K, M, and V. . No participants appear to have wildly-differing accuracy among any of the index finger combinations. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING E2 Left Index ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● S2 Left Index ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Right Index Right Index Right Index Right Index ● Right Index Right Index ● ● Left and Right Index S1 Left Index Left and Right Index Q2 Left Index Left and Right Index Q1 Left Index Left and Right Index P2 Left Index Left and Right Index P1 Left Index Right Index ● Right Index ● ● Right Index ● Right Index Right Index Right Index ● ● Left and Right Index O2 Left Index Left and Right Index O1 Left Index Left and Right Index M2 Left Index Left and Right Index M1 Left Index Left and Right Index L2 Left Index Left and Right Index L1 Left Index Right Index ● Right Index ● Right Index Right Index Right Index Right Index ● ● Left and Right Index K2 Left Index Left and Right Index K1 Left Index Left and Right Index J2 Left Index Left and Right Index J1 Left Index Left and Right Index I2 Left Index Left and Right Index I1 Left Index Right Index ● ● Right Index ● ● Right Index ● ● Right Index Right Index ● ● Left and Right Index H2 Left Index Left and Right Index H1 Left Index Left and Right Index G2 Left Index Left and Right Index G1 Left Index Left and Right Index F2 Left Index Right Index ● F1 Right Index Left and Right Index ● ● ● Left Index ● Left and Right Index ● ● ● Right Index ● ● Right Index ● ● Right Index Right Index ● ● Left and Right Index E1 Left Index Left and Right Index D2 Left Index Left and Right Index D1 Left Index Left and Right Index C2 Left Index Left and Right Index C1 Left Index Right Index Left and Right Index 200 T1 T2 U1 U2 V1 V2 Left Index Left Index Left Index Left Index Left Index Left Index ● ● ● Left and Right Index ● Left and Right Index ● ● ● ● Right Index ● Left and Right Index ● Right Index Left and Right Index ● ● Right Index Left and Right Index ● ● Right Index ● ● Right Index ● Right Index Left and Right Index ● Figure 108: Star plots of relative accuracy for Class A. Values plotted along each radius are log10 of values from Tables 7 through 9, scaled independently from 0 to 1. Values printed as NA in those tables were set to 1 prior to scaling. 30 000 searches were run against an enrollment set of 100 000 subjects for left index and right index, and 1 600 000 subjects for left and right index. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING K.2 201 Class B Relative FNIR results for left slap, right slap, left and right slap, and IDFlat searches are shown in Figure 109. Some notable observations include: . The most accurate submissions (I and Q) are instantly recognizable by their small shape. . The most accurate submissions are also square, meaning they perform equally well for all finger combinations. . Most submissions appear to skew towards slightly worse accuracy on left slap and right slap searches, as indicated by the obtuse angles at the vertices for IDFlats and left and right slap. . Some submissions, like C, F, H, M, and S, vary significantly as the number of fingers at their disposal changes. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology E2 Left Slap ● ● ● ● ● ● ● ● ● ● ● ● Left and Right Slap G1 G2 H1 H2 Left Slap Left Slap Left Slap Left Slap Left Slap Left Slap ● ● ● ● Left and Right Slap ● ● Left and Right Slap ● ● ● Identification Flats ● Left and Right Slap ● ● ● Left and Right Slap L2 Left Slap ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Q1 Q2 Left Slap Left Slap Left Slap Left Slap Left Slap Left Slap ● ● Left and Right Slap ● ● ● Left and Right Slap ● ● ● ● Left and Right Slap Right Slap Left and Right Slap ● ● Right Slap ● ● ● Right Slap Left and Right Slap ● Right Slap ● ● Right Slap ● Right Slap ● ● Left and Right Slap Identification Flats Left and Right Slap O2 Identification Flats Left and Right Slap O1 Identification Flats Left and Right Slap M2 Identification Flats Left and Right Slap M1 Identification Flats Left and Right Slap ● ● ● ● Left and Right Slap V2 Left Slap ● ● ● Left and Right Slap ● ● ● Left and Right Slap ● ● ● ● Left and Right Slap ● ● ● ● Right Slap Left and Right Slap ● Right Slap ● ● Right Slap Left and Right Slap ● Right Slap ● ● Right Slap ● Right Slap ● ● Identification Flats V1 Left Slap Identification Flats U2 Left Slap Identification Flats U1 Left Slap Identification Flats S2 Left Slap Identification Flats S1 Left Slap ● Right Slap ● Right Slap ● ● ● Right Slap ● Right Slap ● ● Identification Flats L1 Left Slap Identification Flats J2 Left Slap Identification Flats J1 Left Slap Identification Flats I2 Left Slap Identification Flats I1 Left Slap Right Slap Identification Flats Left and Right Slap ● ● Right Slap ● ● Left and Right Slap Right Slap ● ● Right Slap ● ● Right Slap ● Right Slap ● ● Identification Flats Left and Right Slap F2 Identification Flats Left and Right Slap F1 Identification Flats ● Left and Right Slap Identification Flats ● Right Slap Identification Flats ● ● ● Left and Right Slap Left and Right Slap Identification Flats ● Right Slap ● ● Right Slap ● Right Slap ● Right Slap ● Right Slap ● Identification Flats E1 Left Slap Identification Flats D2 Left Slap Identification Flats D1 Left Slap Identification Flats C2 Left Slap Identification Flats C1 Left Slap Right Slap Identification Flats FPVTE – F INGERPRINT M ATCHING Right Slap Identification Flats 202 Left and Right Slap Figure 109: Star plots of relative accuracy for Class B. Values plotted along each radius are log10 of values from Tables 10 through 13, scaled independently from 0 to 1. Values printed as NA in those tables were set to 1 prior to scaling. 30 000 searches were run against an enrollment set of 3 000 000 subjects. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING K.3 203 Class C Relative FNIR results for ten-finger plain-to-plain, ten-finger rolled-to-rolled, and ten-finger plain-to-rolled searches are shown in Figure 110. Some notable observations include: . The most accurate submissions (D, I, and Q) are instantly recognizable by their small shape and are seemingly equilateral, indicating that they perform equally well for all finger combinations. . Many submissions had lower accuracy with plain-to-rolled searching over searching homogeneous impressions, including F, G, and M. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 204 FPVTE – F INGERPRINT M ATCHING C1 C2 D1 D2 E1 E2 Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain ● ● ● Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled F2 G1 G2 H1 H2 Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain ● ● ● ● ● Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled ● ● ● ● I1 I2 J1 J2 L1 L2 Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled ● ● Plain−to−Rolled F1 Plain−to−Rolled Plain−to−Rolled ● ● Plain−to−Plain ● Plain−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled Rolled−to−Rolled ● ● Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled ● ● ● ● ● Plain−to−Rolled ● ● ● Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled M1 M2 O1 O2 Q1 Q2 Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain ● ● ● ● Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled ● ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● Rolled−to−Rolled ● Rolled−to−Rolled Rolled−to−Rolled ● Rolled−to−Rolled ● ● ● Plain−to−Rolled Plain−to−Rolled ● ● Plain−to−Plain ● Plain−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● Rolled−to−Rolled ● Rolled−to−Rolled Rolled−to−Rolled ● ● ● Plain−to−Rolled ● ● ● ● S1 S2 U1 U2 V1 V2 Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain Plain−to−Plain ● ● ● ● ● Plain−to−Rolled ● Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled Plain−to−Rolled ● ● ● Rolled−to−Rolled ● ● ● Rolled−to−Rolled ● Rolled−to−Rolled ● Rolled−to−Rolled ● Rolled−to−Rolled ● Rolled−to−Rolled Plain−to−Rolled ● ● Figure 110: Star plots of relative accuracy for Class C. Values plotted along each radius are log10 of values from Tables 14 through 16, scaled independently from 0 to 1. Values printed as NA in those tables were set to 1 prior to scaling. 30 000 searches were run against an enrollment set of 5 000 000 subjects. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashedgray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING K.4 205 All Classes Relative FNIR results for left and right index, IDFlat, and ten-finger rolled-to-rolled searches are shown in Figure 111. Note that the enrollment set sizes differed between all three finger searching combinations. Some notable observations include: . The top-performing submissions from Class A, B, and C (D, I, Q, and V) are instantly recognizable by their small shape. . Many of the most accurate submissions from all three classes appear equally as accurate in each individual class. . If a submission had trouble with a particular class in FpVTE, it appears to be Class C, as seen by the long line segments extending from the ten-finger rolled-to-rolled vertex in most plots that do not contain a mostly equilateral triangle. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 206 FPVTE – F INGERPRINT M ATCHING ● ● ● ● ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● ● Rolled−to−Rolled ● Rolled−to−Rolled ● ● ● G1 G2 H1 H2 Left and Right Index Left and Right Index Left and Right Index Left and Right Index ● ● ● ● ● Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled ● ● ● ● I1 I2 J1 J2 L1 L2 Left and Right Index Left and Right Index Left and Right Index Left and Right Index Left and Right Index Left and Right Index ● ● ● ● ● ● Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled ● ● ● Q1 Q2 Left and Right Index Left and Right Index ● ● ● ● ● ● Rolled−to−Rolled ● Identification Flats ● Identification Flats ● Identification Flats ● ● Identification Flats ● Identification Flats ● Identification Flats ● Rolled−to−Rolled O2 Left and Right Index Rolled−to−Rolled O1 Left and Right Index ● Rolled−to−Rolled M2 Left and Right Index ● Rolled−to−Rolled M1 Left and Right Index ● ● S1 S2 U1 U2 V1 V2 Left and Right Index Left and Right Index Left and Right Index Left and Right Index Left and Right Index Left and Right Index ● ● ● Identification Flats ● Identification Flats ● ● Identification Flats ● ● Identification Flats ● ● Identification Flats Identification Flats ● Identification Flats ● ● Identification Flats ● ● Identification Flats Identification Flats ● ● Rolled−to−Rolled ● Identification Flats Identification Flats ● ● Rolled−to−Rolled F2 Left and Right Index Rolled−to−Rolled F1 Left and Right Index ● Rolled−to−Rolled Identification Flats E2 Left and Right Index Identification Flats E1 Left and Right Index Identification Flats D2 Left and Right Index Identification Flats D1 Left and Right Index Identification Flats C2 Left and Right Index Identification Flats C1 Left and Right Index Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled Rolled−to−Rolled ● ● ● ● Identification Flats ● ● ● Identification Flats ● ● Identification Flats ● ● Identification Flats ● ● Identification Flats ● Identification Flats Rolled−to−Rolled ● Figure 111: Star plots of relative accuracy for two-index finger, IDFlat, and ten-finger rolled-to-rolled comparisons. Values plotted along each radius are log10 of values from Tables 9, 13, and 15, scaled independently from 0 to 1. Values printed as NA in those tables were set to 1 prior to scaling. 30 000 searches were run against an enrollment set of 1 600 000 subjects for left and right index, 3 000 000 subjects for IDFlats, and 5 000 000 subjects for ten-finger rolled-to-rolled. The intersection point of a radius and the circumference of a circle indicate the scaled values 0, 0.5, and 1, with 1 resting on the outer dashed-gray circle. The title above each plot represents the participant’s letter code found on the footer of this page and an identifier used to differentiate between the two submissions each participant could make. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING L 207 Combined Sorted Rankings for a Theoretical Use Case In prior sections, numerous tables and plots were presented showcasing individual statistics of participant submissions. Tables 91 through 98 attempt to coalesce all the presented identification statistics and categorize the overall performance of the submissions. Note that these rankings are theoretical and do not imply quality or endorsement of any kind. Please read the disclaimer for more information. For each statistic, the values and overall ranks from Sections 7 through 9 are displayed. The Score column is computed by weighting percentiles of the results per class. The weights and percentiles were chosen based on values that provided reasonable and consistent results across all enrollment sets for all classes. . Search Template Enrollment Time +2B ⇐⇒ Vf ≤ P10 or +b ⇐⇒ Vf ≤ P15 −b ⇐⇒ Vf ≥ P85 +b ⇐⇒ Vf ≤ P15 −2B ⇐⇒ Vf ≥ P85 or . FNIR +(N − RFNIR ) +5B ⇐⇒ . RAM Consumption +2b ⇐⇒ Vf ≤ P10 or or . Identification Time +(N − RTime ) V f ≤ P5 or +2B ⇐⇒ Vf ≤ P10 or +3B ⇐⇒ Vf ≤ P15 or +B ⇐⇒ Vf ≤ P15 or +B ⇐⇒ V f ≤ P5 or −2B ⇐⇒ Vf ≥ P90 or −5B ⇐⇒ Vf ≥ P95 or −B ⇐⇒ Vf ≥ P85 −3B ⇐⇒ Vf ≥ P90 or −B ⇐⇒ Vf ≥ P75 where N = number of participants for finger B = max (N/2.0 , 15) b = max (N/6.0 , 3) Rf = submission’s rank for factor f Vf = submission’s value for factor f Px = xth percentile of Vf As shown, the primary factors used in scoring were accuracy and identification speed. While enrollment speed is important, the FpVTE API limited the time that could be used, and most participants completed enrollment in a relatively similar timeframe. Using a very high amount of RAM was penalized heavily, as greater amounts of RAM require additional compute nodes, while less RAM could allow additional identification processes to be run, decreasing search time. High accuracy was rewarded the most of any factor—regardless of how fast or resource-friendly a submission is, it doesn’t matter if the results are incorrect. The color bands are indicative of the 80th percentile, 55th percentile, and below. The best performance in each category is shaded in green and the worst in pink. Time values in the Identification and Enroll columns are reported in seconds, but were originally recorded to microsecond precision. RAM is the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method. Some notable observations from Tables 90 through 98 include: . Some of the most accurate submissions used such a small amount of RAM that only a single compute node was necessary. . The most accurate submissions did not use the entire 90 seconds to perform identifications. This dramatically reduced the score given to submissions that used the maximum amount of time but were not with the top accuracy range. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 208 FPVTE – F INGERPRINT M ATCHING . In Class C, there are a number of submissions that achieve an “acceptable” level of accuracy in much less time than the most accurate submissions. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Score FNIR Letter Sub. # D 1 109 1 0.0197 Identification 20 7.32 209 Enrollment RAM 29 1.43 19 0.61 L 1 95 8 0.0625 2 0.29 1 0.05 26 1.14 V 1 82 4 0.0253 19 6.01 10 0.32 17 0.39 Q 2 79 3 0.0226 24 10.43 19 0.54 14 0.34 Q 1 74 2 0.0222 28 15.70 19 0.54 13 0.34 I 1 73 5 0.0257 22 9.94 30 1.46 20 0.70 J 1 73 14 0.0786 4 0.54 11 0.32 9 0.32 O 1 60 15 0.0818 5 0.62 4 0.27 15 0.35 E 1 59 12 0.0745 3 0.35 13 0.35 25 1.12 L 2 57 6 0.0351 12 3.32 3 0.15 28 2.32 C 1 50 24 0.1335 1 0.26 17 0.39 7 0.18 O 2 43 13 0.0766 9 1.56 4 0.27 15 0.35 T 2 43 9 0.0685 18 5.97 8 0.31 1 0.01 J 2 43 10 0.0712 7 1.25 11 0.32 9 0.32 F 2 27 18 0.1082 14 3.56 25 1.07 5 0.18 P 2 25 22 0.1272 13 3.33 15 0.36 22 0.78 F 1 24 20 0.1111 11 2.49 25 1.07 6 0.18 G 1 20 19 0.1089 21 9.74 21 0.55 18 0.57 C 2 14 25 0.1337 6 0.76 17 0.39 8 0.18 P 1 14 23 0.1308 8 1.32 15 0.36 23 0.78 S 1 10 7 0.0571 29 16.86 22 0.88 21 0.72 H 1 5 26 0.1576 15 3.61 6 0.30 11 0.33 H 2 1 27 0.1607 17 4.13 6 0.30 11 0.33 U 1 −2 21 0.1218 27 14.42 2 0.15 27 1.70 K 2 −10 16 0.0875 23 10.32 23 1.05 30 3.87 11 0.0723 30 1.11 E 2 −11 16.90 13 0.35 24 K 1 −12 17 0.0883 25 10.44 23 1.05 29 3.87 M 1 −49 29 0.2995 10 1.84 27 1.07 3 0.14 T 1 −51 30 NA 16 3.99 8 0.31 2 0.01 M 2 −55 28 0.2921 26 10.70 27 1.07 4 0.14 D 2 84 1 0.0197 3 41.84 5 1.42 3 0.61 S 2 4 4 0.0650 4 44.66 4 0.88 4 0.72 I 2 3 3 0.0278 2 40.99 6 2.15 5 1.04 V 2 −5 2 0.0252 6 65.35 2 0.32 1 0.39 G 2 −13 5 0.1086 5 54.03 3 0.55 2 0.57 U 2 −40 6 0.1178 1 24.16 1 0.15 6 1.42 Table 90: Tabulation of operationally-ranked results for Class A — Left Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 100 000, as seen in Table 17. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 72. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 60. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 210 FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Score FNIR Identification Enrollment Letter Sub. # Q 2 111 2 0.0214 23 9.98 19 D 1 109 1 0.0190 20 7.46 L 1 95 8 0.0505 2 0.26 V 1 81 5 0.0223 18 E 1 76 11 0.0630 3 RAM 0.54 11 0.33 29 1.43 19 0.61 1 0.05 26 1.14 5.60 10 0.32 17 0.39 0.32 13 0.36 24 1.07 I 1 74 3 0.0215 22 9.83 30 1.46 20 0.69 O 1 58 17 0.0776 5 0.56 4 0.27 15 0.34 L 2 57 6 0.0295 13 3.26 3 0.16 28 2.32 J 1 55 16 0.0708 4 0.51 11 0.33 10 0.31 C 1 50 24 0.1132 1 0.24 17 0.39 7 0.17 Q 1 43 4 0.0218 28 14.24 19 0.54 11 0.33 O 2 43 13 0.0675 9 1.36 4 0.27 16 0.34 T 2 43 9 0.0562 19 5.87 8 0.31 1 0.01 J 2 41 12 0.0643 7 1.13 11 0.33 9 0.31 F 2 27 18 0.0903 15 3.60 25 1.09 5 0.17 P 2 25 22 0.1100 12 3.07 15 0.37 23 0.76 F 1 24 20 0.0933 11 2.38 25 1.09 6 0.17 G 1 20 19 0.0910 24 10.10 21 0.55 18 0.57 C 2 16 23 0.1124 6 0.69 17 0.39 8 0.18 P 1 12 25 0.1133 8 1.24 15 0.37 22 0.76 S 1 9 7 0.0442 30 16.55 22 0.88 21 0.71 H 1 5 26 0.1230 14 3.56 6 0.30 14 0.33 H 2 1 27 0.1249 17 4.05 6 0.30 13 0.33 U 1 −3 21 0.0996 27 10.76 2 0.15 27 1.56 K 2 −9 15 0.0685 25 10.27 23 1.05 29 3.87 K 1 −9 14 0.0682 26 10.42 23 1.05 30 3.87 E 2 −10 10 0.0624 29 16.27 13 0.36 25 1.07 T 1 −19 28 0.1929 16 3.73 8 0.31 1 0.01 M 1 −50 30 0.2615 10 1.78 27 1.09 3 0.14 M 2 −80 29 0.2526 21 9.39 27 1.09 3 0.14 D 2 84 1 0.0190 2 28.64 5 1.43 3 0.61 I 2 19 2 0.0214 3 40.35 6 2.15 5 1.01 S 2 4 4 0.0503 4 46.05 4 0.88 4 0.71 G 2 −13 5 0.0909 5 59.55 3 0.55 2 0.57 V 2 −21 3 0.0222 6 61.24 2 0.32 1 0.39 U 2 −40 6 0.1007 1 20.10 1 0.15 6 1.30 Table 91: Tabulation of operationally-ranked results for Class A — Right Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 100 000, as seen in Table 17. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 73. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 61. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING ≥ 20 seconds < 20 seconds Participant Score FNIR 211 Identification Enrollment 0.0034 5 9.29 3 0.64 5 11.77 4 0.0146 2 2.19 1 0.11 8 18.76 3 0.0143 6 8.30 Letter Sub. # V 1 87 1 L 1 43 RAM J 1 24 13.64 4 0.65 3 I 1 19 2 0.0058 8 17.87 9 2.93 6 15.83 C 1 12 7 0.0368 1 2.08 5 0.78 2 4.87 O 1 9 5 0.0229 7 14.46 2 0.53 4 9.05 C 2 −3 8 0.0374 4 6.35 5 0.78 1 4.87 K 1 −44 6 0.0360 9 18.01 8 2.09 9 61.81 G 1 −69 9 0.0515 3 5.22 7 1.09 7 16.38 Q 2 109 1 0.0027 19 161.02 15 1.08 9 9.14 Q 1 107 1 0.0027 21 212.69 15 1.08 10 9.14 L 2 89 7 0.0072 3 23.42 3 0.31 26 53.98 V 2 78 3 0.0028 18 127.65 9 0.64 13 11.77 D 1 74 4 0.0030 15 70.99 25 2.84 17 18.58 E 1 71 11 0.0207 1 16.35 11 0.70 22 33.40 S 1 70 13 0.0281 2 23.00 18 1.75 14 14.82 D 2 68 4 0.0030 23 237.43 26 2.84 17 18.58 J 2 39 8 0.0143 7 33.35 10 0.65 7 8.30 O 2 36 12 0.0214 10 43.53 4 0.53 8 9.05 I 2 36 4 0.0030 25 338.88 27 4.30 21 30.83 T 2 35 19 0.0366 9 37.23 7 0.62 1 0.01 U 1 26 17 0.0336 11 45.51 1 0.30 25 44.63 P 2 21 16 0.0333 17 101.16 13 0.72 19 21.41 P 1 20 20 0.0370 14 63.65 13 0.72 20 21.42 G 2 17 15 0.0311 22 221.16 17 1.09 16 16.38 H 1 7 24 0.0686 8 36.71 5 0.60 12 9.36 F 1 5 21 0.0386 13 60.66 21 2.15 5 4.86 K 2 4 14 0.0286 6 32.93 20 2.09 27 61.81 H 2 4 23 0.0684 12 52.25 5 0.60 11 9.36 F 2 2 22 0.0412 16 73.95 21 2.15 5 4.86 U 2 −3 18 0.0358 24 240.40 1 0.30 24 37.35 S 2 −11 9 0.0195 26 495.50 18 1.75 14 14.82 E 2 −13 10 0.0202 27 518.11 11 0.70 23 33.41 T 1 −28 27 NA 4 25.68 7 0.62 1 0.01 M 2 −36 25 NA 20 171.24 23 2.16 4 3.76 M 1 −48 26 NA 5 31.44 23 2.16 3 3.75 Table 92: Tabulation of operationally-ranked results for Class A — Left and Right Index. Submissions were split into two groups. The first group includes submissions that performed searches on average in less than 20 seconds, and the second includes those that took, on average, 20 seconds or longer. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 1 600 000, as seen in Table 18. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 74. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 62. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 212 FPVTE – F INGERPRINT M ATCHING Participant Score FNIR Identification Enrollment RAM Letter Sub. # Q 1 126 2 0.0098 19 53.24 20 3.41 1 7.54 I 2 111 1 0.0094 13 38.26 30 7.67 21 108.71 L 1 101 16 0.0288 3 5.56 2 1.11 22 119.79 Q 2 96 3 0.0099 20 54.02 20 3.41 2 7.54 I 1 83 4 0.0116 21 55.96 25 6.50 5 49.68 V 2 83 8 0.0190 15 47.97 7 1.22 9 63.53 D 2 79 5 0.0142 22 58.10 24 4.17 12 79.43 C 1 76 22 0.0654 2 3.74 12 2.12 3 46.49 E 1 76 13 0.0259 1 2.82 11 1.95 14 79.72 V 1 71 9 0.0192 11 36.07 7 1.22 8 63.53 L 2 56 14 0.0276 5 12.37 1 0.33 27 177.49 C 2 55 21 0.0647 4 6.74 12 2.12 4 46.49 D 1 49 6 0.0163 18 52.23 14 2.15 13 79.43 O 1 42 12 0.0257 12 37.01 5 1.21 17 101.00 J 1 38 15 0.0287 7 25.61 3 1.21 20 101.00 O 2 30 11 0.0254 26 73.15 5 1.21 18 101.00 E 2 26 7 0.0187 10 34.37 15 2.21 28 318.03 G 2 21 17 0.0325 23 59.49 18 2.83 25 156.12 J 2 8 10 0.0236 27 74.38 3 1.21 19 101.00 H 1 7 23 0.0998 16 49.75 9 1.67 16 86.46 G 1 6 18 0.0371 6 16.24 18 2.83 26 156.12 H 2 2 24 0.1008 17 51.77 9 1.67 15 86.46 S 1 −5 25 0.1089 24 71.69 22 3.96 24 150.35 S 2 −22 26 0.1133 25 71.77 22 3.96 23 150.35 F 2 −31 28 0.1681 14 43.59 26 6.73 10 77.02 M 2 −47 27 0.1634 29 87.21 28 6.73 7 57.53 U 1 −48 20 0.0500 28 80.03 17 2.42 29 440.67 U 2 −48 19 0.0461 30 89.07 16 2.41 30 540.60 F 1 −57 29 0.1684 8 30.09 26 6.73 11 77.02 M 1 −59 30 0.1736 9 31.17 28 6.73 6 57.53 Table 93: Tabulation of operationally-ranked results for Class B — Left Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 75. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Score FNIR Identification 213 Enrollment RAM Letter Sub. # I 2 113 1 0.0045 14 41.13 30 7.65 21 108.71 D 2 113 2 0.0052 21 57.19 24 4.13 12 79.43 15 0.0167 3 119.78 L 1 102 5.47 2 1.11 22 Q 2 91 3 0.0057 22 61.94 20 3.41 2 7.54 Q 1 90 3 0.0057 23 64.04 20 3.41 1 7.54 V 1 87 8 0.0106 11 35.43 3 1.19 8 63.53 I 1 82 5 0.0058 20 55.87 25 6.47 5 49.68 E 1 75 13 0.0151 1 2.97 11 1.91 14 79.72 V 2 66 9 0.0110 16 48.51 3 1.19 9 63.53 C 1 55 24 0.0403 2 3.67 12 2.09 4 46.49 D 1 50 6 0.0072 19 53.32 14 2.14 13 79.43 J 1 44 14 0.0156 7 24.14 5 1.20 18 101.00 C 2 43 23 0.0392 4 6.50 12 2.09 3 46.49 O 1 37 12 0.0142 12 35.61 7 1.20 17 101.00 L 2 37 17 0.0202 6 12.66 1 0.33 27 177.49 G 2 31 16 0.0198 13 37.28 18 2.78 25 156.12 E 2 26 7 0.0083 10 34.47 15 2.18 28 318.00 G 1 22 18 0.0212 5 10.13 18 2.78 26 156.12 S 2 14 22 0.0381 25 70.84 22 3.88 24 150.35 S 1 14 21 0.0369 24 70.54 22 3.88 23 150.35 J 2 13 10 0.0126 27 74.09 5 1.20 19 101.00 O 2 8 11 0.0132 26 73.93 7 1.20 20 101.00 H 1 3 25 0.0641 17 50.16 9 1.63 16 86.46 H 2 1 26 0.0647 18 52.84 9 1.63 15 86.46 F 2 −32 28 0.1220 15 45.92 26 6.68 10 77.02 M 2 −47 27 0.1155 29 87.41 28 6.69 6 57.53 U 1 −48 19 0.0266 30 89.26 16 2.39 29 440.67 U 2 −48 20 0.0273 28 80.11 17 2.39 30 540.59 F 1 −57 29 0.1222 8 31.87 26 6.68 11 77.02 M 1 −59 30 0.1259 9 32.03 28 6.69 7 57.53 Table 94: Tabulation of operationally-ranked results for Class B — Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 76. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 214 FPVTE – F INGERPRINT M ATCHING Participant Score FNIR Identification Enrollment RAM Letter Sub. # Q 1 123 2 0.0021 22 65.02 20 6.84 2 7.54 I 2 110 1 0.0015 17 46.13 30 15.33 21 108.71 Q 2 94 3 0.0022 19 53.40 20 6.84 1 7.54 I 1 92 3 0.0022 13 37.19 25 12.96 5 49.68 L 1 89 12 0.0054 4 10.48 2 2.23 22 119.77 V 1 87 7 0.0036 15 38.93 7 2.40 9 63.53 V 2 81 7 0.0036 18 52.80 7 2.40 8 63.53 D 2 76 5 0.0024 21 63.05 24 8.28 13 79.43 E 1 72 15 0.0063 2 6.12 11 3.85 14 79.72 D 1 48 6 0.0031 20 61.24 14 4.28 12 79.43 G 1 41 18 0.0106 1 3.89 18 5.62 26 156.12 L 2 40 14 0.0062 6 20.78 1 0.66 27 177.49 O 1 39 13 0.0057 14 38.34 3 2.40 20 101.00 J 1 37 16 0.0068 7 26.15 5 2.40 18 101.00 O 2 29 11 0.0051 23 68.46 3 2.40 19 101.00 G 2 29 17 0.0084 11 34.55 18 5.62 25 156.12 J 2 10 9 0.0047 24 69.28 5 2.40 17 101.00 E 2 8 10 0.0049 10 33.09 15 4.38 28 317.99 M 1 4 27 0.0904 8 27.15 28 13.53 7 57.53 F 2 4 26 0.0901 12 36.36 26 13.49 11 77.02 H 1 0 23 0.0349 25 70.38 9 3.31 16 86.46 M 2 −1 25 0.0882 16 45.91 28 13.53 6 57.53 H 2 −2 24 0.0361 26 73.06 9 3.31 15 86.46 22 0.0190 27 150.35 S 2 −4 82.13 22 7.83 24 C 1 −7 30 NA 3 6.40 12 4.20 3 46.49 S 1 −19 21 0.0160 29 83.33 22 7.83 23 150.35 F 1 −26 28 0.0910 9 27.27 26 13.49 10 77.02 C 2 −29 29 NA 5 10.70 12 4.20 4 46.49 U 1 −49 20 0.0139 30 83.62 16 4.82 29 440.67 U 2 −49 19 0.0124 28 82.40 17 4.83 30 540.59 Table 95: Tabulation of operationally-ranked results for Class B — Left and Right Slap. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 77. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Score FNIR 215 Identification Enrollment 0.0012 7 19.07 25 16.52 5 49.68 2 0.0012 14 42.92 20 8.92 1 7.54 2 0.0012 24 66.01 20 8.92 2 7.54 2 0.0012 18 46.70 24 11.86 13 79.43 1 0.0009 16 43.86 30 19.36 21 108.71 Letter Sub. # I 1 132 2 Q 1 129 Q 2 119 D 2 114 I 2 109 RAM L 1 90 10 0.0031 5 14.50 2 3.05 22 119.79 V 2 83 7 0.0024 19 49.30 7 3.36 9 63.53 E 1 72 16 0.0043 2 6.76 13 5.34 14 79.73 G 1 71 18 0.0062 1 4.26 18 7.90 25 156.12 V 1 71 9 0.0027 10 34.96 7 3.36 8 63.53 D 1 52 6 0.0020 17 45.15 17 6.16 12 79.43 L 2 41 11 0.0033 9 28.59 1 0.88 27 177.48 O 1 39 15 0.0041 11 35.64 5 3.36 19 101.00 J 1 36 17 0.0049 8 25.38 3 3.35 18 101.00 O 2 31 13 0.0035 21 60.02 5 3.36 20 101.00 J 2 27 11 0.0033 22 60.13 3 3.35 17 101.00 E 2 20 7 0.0024 15 43.61 16 5.93 28 317.95 G 2 7 14 0.0040 6 16.26 18 7.90 26 156.12 M 1 1 26 0.0543 13 38.87 28 18.11 6 57.53 M 2 −8 25 0.0515 23 65.91 28 18.11 6 57.53 C 1 −8 30 NA 3 7.92 11 5.13 3 46.49 U 2 −17 22 0.0141 25 75.28 14 5.73 30 540.60 S 1 −18 20 0.0108 29 87.60 22 10.31 24 150.35 H 1 −19 23 0.0203 26 82.66 9 4.37 15 86.46 S 2 −20 21 0.0136 30 88.46 22 10.31 23 150.35 H 2 −21 24 0.0204 27 86.56 9 4.37 16 86.46 C 2 −28 29 NA 4 10.21 11 5.13 4 46.49 F 1 −29 27 0.0591 12 37.56 26 18.07 10 77.02 F 2 −33 27 0.0591 20 49.34 26 18.07 10 77.02 U 1 −49 19 0.0099 28 86.60 15 5.82 29 440.68 Table 96: Tabulation of operationally-ranked results for Class B — Identification Flats. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 3 000 000, as seen in Table 19. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 78. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 63. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology 216 FPVTE – F INGERPRINT M ATCHING Participant Score FNIR Letter Sub. # Q 1 133 2 Identification Enrollment RAM 0.0011 9 36.86 20 9.11 2 13.33 I 2 114 1 0.0010 10 38.28 30 16.99 11 110.22 D 2 108 2 0.0011 26 74.36 24 13.66 15 132.37 Q 2 102 4 0.0013 8 35.35 20 9.11 1 13.33 L 1 101 16 0.0102 3 17.51 1 3.48 23 192.19 I 1 78 4 0.0013 16 52.41 29 16.67 4 84.42 E 1 74 14 0.0088 2 12.77 11 5.20 14 127.39 J 1 66 12 0.0047 12 43.57 5 3.62 20 181.00 V 1 55 7 0.0024 14 47.02 7 3.66 13 121.80 G 1 51 23 0.0368 1 7.71 18 8.32 25 274.35 V 2 51 7 0.0024 19 62.26 7 3.66 12 121.79 L 2 39 15 0.0095 6 23.53 1 3.48 27 274.95 O 1 35 9 0.0025 23 68.54 3 3.61 19 181.00 C 2 32 24 0.0711 5 18.38 13 5.80 10 92.86 O 2 31 10 0.0027 24 73.41 3 3.61 22 181.00 H 2 23 19 0.0275 20 65.36 9 5.16 18 144.02 H 1 21 20 0.0276 21 68.07 9 5.16 17 144.02 D 1 11 6 0.0015 25 73.93 17 7.10 16 132.37 J 2 9 10 0.0027 28 77.75 6 3.62 21 181.00 G 2 3 20 0.0276 7 23.69 18 8.32 26 274.35 M 2 2 27 0.0826 13 46.70 25 15.83 8 90.93 F 1 1 25 0.0734 15 51.69 27 15.87 7 90.93 F 2 −1 25 0.0734 18 60.86 27 15.87 6 90.93 U 2 −7 17 0.0155 22 68.30 15 5.91 29 811.45 U 1 −10 18 0.0163 27 74.94 15 5.91 29 811.45 S 1 −21 22 0.0311 29 86.56 22 10.87 24 260.37 M 1 −21 28 0.0934 11 43.37 25 15.83 5 90.93 E 2 −25 13 0.0048 17 52.91 12 5.20 28 573.25 C 1 −35 30 NA 4 18.30 13 5.80 9 92.86 S 2 −94 29 0.1680 30 91.74 22 10.87 3 61.04 Table 97: Tabulation of operationally-ranked results for Class C — Ten-Finger Plain-to-Plain. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . NA indicates that the operations required to produce the value could not be performed. The Identification column shows the time used to perform a search over an enrollment set of 5 000 000, as seen in Table 20. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 79. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 64. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology FPVTE – F INGERPRINT M ATCHING Participant Score FNIR 217 Identification Enrollment 0.0014 7 30.82 24 18.73 11 137.03 1 0.0013 15 54.56 25 20.23 4 113.80 2 101 2 0.0014 25 74.22 17 13.06 2 20.22 1 90 5 0.0017 10 40.74 9 8.50 17 234.03 D 1 84 4 0.0015 17 58.92 21 17.39 10 132.40 L 2 73 14 0.0083 5 20.25 3 4.36 26 367.82 E 1 71 18 0.0106 2 8.63 12 9.92 16 191.66 Q 1 69 5 0.0017 26 74.40 17 13.06 1 20.22 L 1 66 17 0.0097 8 31.68 3 4.36 21 280.49 C 2 55 15 0.0085 6 21.97 13 10.69 15 183.36 C 1 55 16 0.0094 4 20.15 13 10.69 14 183.36 Letter Sub. # I 2 124 2 I 1 115 Q V RAM V 2 51 8 0.0019 18 65.07 9 8.50 18 234.03 G 1 49 24 0.0447 1 7.78 19 15.50 20 261.29 J 1 44 13 0.0051 9 33.02 5 6.69 22 303.13 O 1 34 11 0.0034 14 54.42 6 6.71 25 303.13 G 2 33 21 0.0333 3 19.46 19 15.50 19 261.29 O 2 32 9 0.0033 21 69.30 6 6.71 23 303.13 J 2 31 9 0.0033 19 67.98 8 6.74 24 303.13 U 2 15 22 0.0351 24 72.48 1 2.87 28 806.17 M 2 6 27 0.0716 13 48.80 26 20.89 5 130.29 E 2 4 12 0.0050 11 42.73 11 9.91 30 930.48 D 2 3 7 0.0018 27 74.97 30 30.47 9 132.37 F 1 2 25 0.0536 16 55.79 28 20.92 7 130.29 U 1 −3 23 0.0358 28 82.61 1 2.87 28 806.17 F 2 −3 25 0.0536 20 68.20 28 20.92 6 130.29 H 2 −18 19 0.0199 29 84.50 15 12.17 12 144.02 H 1 −18 20 0.0201 30 84.51 15 12.17 13 144.02 M 1 −29 28 0.0783 12 48.08 26 20.89 8 130.29 S 2 −56 30 0.2462 23 70.77 22 17.57 3 78.28 S 1 −96 29 0.0860 22 69.71 22 17.57 27 382.88 Table 98: Tabulation of operationally-ranked results for Class C — Ten-Finger Rolled-to-Rolled. Letter refers to the participant’s letter code found on the footer of this page. Sub. # is an identifier used to differentiate between the two submissions each participant could make. Score refers to the sum of the scoring equations shown in Appendix L. The color bands are indicative of the 80th percentile, 55th percentile, and below of the Score column. The FNIR column was computed at the score threshold that gave FPIR = 10−3 . The Identification column shows the time used to perform a search over an enrollment set of 5 000 000, as seen in Table 20. The Enrollment column shows the time used to create a search template to be used for a query, as seen in Table 80. Identification and Enrollment durations are reported in seconds, but were originally recorded to microsecond precision. RAM refers to the sum of the resident set sizes of the stage one identification processes over all compute nodes after returning from the identification stage one initialization method, as seen in Table 65. RAM is reported in gigabytes, where 1 GB is equal to 1 073 741 824 bytes. The number to the left of a value provides the value’s column-wise ranking, with the best performance shaded in green and the worst in pink. The table is sorted on descending Score. C = afis team D = 3M Cogent E = Neurotechnology F = Papillon G = Dermalog H = Hisign Bio-Info Institute I = NEC J = Sonda K = Tiger IT L = Innovatrics M = SPEX O = ID Solutions P = id3 Q = Morpho S = Decatur Industries T = BIO-key U = Aware V = AA Technology