300 likes | 664 Views
MINEX II. An evaluation of fingerprint Match-on-Card technology Patrick Grother Biometrics 2007, London, October 18, 2007. Overview. MINEX II – Match-on-card Compact iris interoperability test Standards for multimodal biometrics NIST Biometric Quality Workshop.
E N D
MINEX II An evaluation of fingerprint Match-on-Card technology Patrick Grother Biometrics 2007, London, October 18, 2007
Overview • MINEX II – Match-on-card • Compact iris interoperability test • Standards for multimodal biometrics • NIST Biometric Quality Workshop
MINEX II – The NIST Context NIST Biometric Testing FRVT (face) 1:N Fingerprint ICE (iris) Quality Data for Credentials FpVTE (2003) US Gov. Systems ELFT (latent) Slap Segmentation PFT (ongoing)
MINEX II – The NIST Context NIST Support for Biometric Elements for Identity Credentials sBMOC MINEX Compact Iris Standards SC37 WG3 MINEX I 2004 Initial evaluation Ongoing MINEX PIV MINEX III Minutia quality calibration MINEX II Match-on-Card
Template Generators Cogent Systems Dermalog Identification Systems Bioscrypt Sagem Morpho Neurotechnologija Innovatrics NEC Cross Match Technologies L1 / Identix Precise Biometrics XTec SecuGen BIO-key International Motorola Aware Sonda Technologies Matchers Cogent Systems Dermalog Identification Systems Bioscrypt Sagem Morpho Neurotechnologija Innovatrics NEC L1 / Identix XTec SecuGen BIO-key International Motorola Aware Startek Engineering Ongoing MINEX Compliant and Eligible for GSA Certification 16 suppliers 14 suppliers
MINEX II – Why MOC? • Match-on-Card – Why • Cards are ubiquitous • ISO/IEC 7816 cards have been 140-2 certified • No central database • Biometric reference never leaves the card • Match-on-Card – Why not? • Verification template must be made off card • And passed to the card • A matcher on every credential • Computational resources …
MINEX II – Why? • Hypothesis: MOC implementations have same accuracy • Why might that be? • MOC is not new. • Same companies are involved • Why not? • Limited computational resources • Stack space, registers • Integer arithmetic • Smaller instruction sets • Smaller templates • MOC typically uses fewer minutiae • Reduced angular resolution in ISO-CC format • Asymmetric Algorithms MINEX II is intended of as a definitive, public, independent, simultaneous measurement of the algorithmic accuracy and speed of MOC implementations
Not in MINEX II Scope • Card reliability, robustness • Card vulnerability • Security evaluation • System-on-card • Proprietary templates • Business model, economics • Card conformance to 7816-x • Contact vs. contactless
Two NIST programs: MINEX II + sBMOC • Two separate but related programs: • MINEX II • Accuracy and speed of card-based algorithms • Contact: patrick.grother@nist.gov • sBMOC “Secure Biometric Match-on-Card” • Demonstration of secure protocols for biometric authentication. Publication of NISTIR 7452 imminent. • Contact: william.macgregor@nist.gov
MINEX II – Design objectives • Make it: independent, statistically robust, repeatable • NIST • Massive offline archival data • Uniform, standards-based, interface • Measure error rate tradeoffs • Consider FNMR(t) vs. FMR(t) Need matcher scores from card • Demonstrate at industry “norm” of FMR of 10-4 • Measure time • Inspect the slow-but-accurate vs. fast-but-inaccurate spectrum • Allow teams • Allow card suppliers to team with fingerprint matcher suppliers • Use the industry-preferred template • ISO/IEC 19794-2 compact card – three bytes per minutia
MINEX II - Schedule • Test plan development • Initiated April 2007, finalized Aug 3, 2007 • Phase I (private) • Submission deadline, September 10, 2007 • Acceptance + Validation testing began September 11, 2007 • Results to vendors October 14 • Phase II (public) • Submissions due late October 2007 • NIST publishes report December 17, 2007 • MINEX II testing protocol standardization • US NB agreed to send New Project Proposal to SC37(WG5)
Authentec Bioscrypt Cogent Daon Fraunhöfer Gemalto IDTP L1 Oberthur Precise Biometrics Sagem SC17 WG11 MINEX II - Acknowledgments • The MINEX test plan established • a definitive card interface for testing • a definitive PC-based interface for testing • profiles of the base minutia standards • was developed in consultation with industry. Thanks to: http://fingerprint.nist.gov/minex/minexII/NIST_MOC_ISO_CC_interop_test_plan_0815.pdf
Evaluation Principle 2. Confirm by repeating n « N comparisons on the card 1: Measure accuracy by Execute N template comparisons on general purpose computer N = O(106) n = O(103)
MINEX II – Execution • Standards based test interface • ISO/IEC 7816-4 – card commands • ISO/IEC 7816-11 – biometric data structures • ISO/IEC 19794-2 – compact card minutiae on card • INCITS 378:2004 – parent template off card • Test protocol • Generate templates on PC • Execute O(106) template comparisons on PC • Repeat selected comparisons on target card • Test on-card and off-card matcher scores for identity
MINEX II – Card APDUs Verification Template sent via VERIFY Reference Template: sent via PUT DATA FNMR FMR Similarity Score via GET DATA
MINEX II - Implementation • Standard hardware • SCR SCM335 reader (contact) • Standard software • M.U.S.C.L.E open-source PC/SC drivers • Linux 2.6.X • NIST Open Source MOC Harness
INCITS 378 as Parents to ISO-CC Template extraction produces INCITS 378 Reader prompts for specific finger Scan produce output image User presents card Reader requests BIT from card Remove N-K minutiae based on quality + polar distance, per BIT Quantize minutia angle (8 6 bits) Quantize (x,y) 197 100 pix cm-1 Sort minutiae (XY, YX, Polar), per BIT ISO/IEC 19794-2 compact card “template” Send to card Match Decision
Remove minutiae to card capacity Strategy: Lowest quality first and, for tied quality values, use largest radial distance.
MINEX II – Guidance on # minutiae FNMR 5 Matchers Fix threshold to give FMR = 0.001 for un-pruned templates FMR Card capacity (max # minutiae)
Does ISO-CC Degrade Accuracy? • ISO/IEC 19794-2 compact card format • ~ 250 dpi (vs. ubiquitous 500) • ~ 5.6 deg. angle resolution (vs. 2 deg in INCITS 378) • FMR decreases slightly (but significantly) • FNMR increases slightly (but significantly)
MINEX II – Software for Biometric Data • Open-source “C” code for • INCITS 378 minutiae • ISO/IEC 19794-2 minutiae • INCITS 385 face (~ ISO/IEC 19794-5) • INCITS 381 finger (~ ISO/IEC 19794-6) • Validation, construction, IO • http://www.itl.nist.gov/iad/894.03/nigos/biomdi.html • Under full version control
MINEX II – Software support for MOC • MOC Template Support • Transcoding INCITS 378 to ISO-CC templates: • http://www.itl.nist.gov/iad/894.03/nigos/biomapp.html • ISO/IEC 7816 Support • MINEX II interface uses (PUT DATA, VERIFY etc) • See http://fingerprint.nist.gov/minexII • And the open-source test driver here • http://www.itl.nist.gov/iad/894.03/nigos/biomapp.html
MINEX II Results • Protocol • Vendor acceptance • Four suppliers • Six implementations • Open source support • It works • One interface problem • Implementations • ISO-CC templates can be matched with accuracy approaching INCITS 378 • Some MOC implementations attain accuracy approaching that of better MINEX 04 matchers • Median VERIFY execution time < 0.5s • Speed – accuracy tradeoff is alive and well, but supplier influence is larger
Compact Iris Formats Compression JPEG 2000 + ROI JPEG Lossless Interoperability • Multiple segmentation algorithms • Multiple matching algorithms NIST will release draft evaluation plan: November 15
Fusion Support • INCITS 439 – Fusion Information Format is about to be published. • It defines binary data structures for similarity score statistics (CDFs) to support simple yet powerful fusion implementations • Multimodal • Multi-algorithm
Score level fusion • Large literature demonstrating that fusion techniques produce lower (FAR,FRR) • If systems behave (fail, succeed) independently then fusion can have maximum effect. • Score-level fusion is more potent that decision level • But some evidence that even (face + finger) and (finger + iris) are partially correlated, due to human-sensor interaction etc. • Score-level fusion is favored over feature level fusion for black box reasons: • Implementation is easy. • Post-match fusion avoids IP licensing or exposure. • Also: • Multimodal: Iris Corp A + Fingerprint Corp B • Multi-algorithmic: Face Corp A + Face Corp B + . . .
Bayes optimal for uncorrelated biometrics Use of likelihood ratio allows relative “strength” of the (two) biometrics comes out in the wash without ad hoc weighting Aka BGI, Neyman Pearson. m(x) n(x) INCITS 439 Fusion Information Format - An Example n(x) m(x) pdf N(x) M(x) cdf = L(x) Fused score: s(x) = log LFACE(xFACE) + log LIRIS(xIRIS) + …
NIST – Biometric Quality Workshop • NIST Biometric Quality Workshop • November 7-8, 2007 • Gaithersburg, MD, USA • Sequel to March 06. • Quality • Uses (during capture) • Relation to error rates • Assessment capabilities • Needs • Interoperable values • Calibration http://www.itl.nist.gov/iad/894.03/quality/workshop07
Feedback is welcome: patrick.grother@nist.gov Thank You MINEX Root http://fingerprint.nist.gov/minex MINEX II http://fingerprint.nist.gov/minexII Ongoing MINEX program http://fingerprint.nist.gov/minex