NIST Activities at the Intersection of Biometrics Standardization and Testing. NIST Support for Match on Card

Transcription

1 NIST Activities at the Intersection of Biometrics Standardization and Testing NIST Support for Match on Card Patrick Grother CTST 2009 May 5, 2009

2 The Minutiae Matching Problem

3 Standardized Templates» INCITS 378:2004» Fingerprint minutiae» A set of (x, y, theta, type, quality) points» Each point in six bytes» Median 38 points per finger» ISO/IEC compact card» (x, y, theta, type)» Each point in 3 bytes Lower spatial and angular resolution» Placement differences Ridge bifurcation (x, y) theta Ridge ending

4 Template Matching Minutiae from enrollment image Minutiae from verification image

5 Biometric Interoperability

6 Cross vendor Interoperability Enrolment sample, (e.g. visa or passport issuance.) Produced by Product A Verification sample (e.g. at border crossing) Produced by Product B STANDARD DATA Matching by Product C STANDARD DATA Decision to admit based on verification decision

7 Cross vendor interoperability Enrolment Template Produced by CrossMatch Authentication Template Produced by Cogent Sagem Matcher» Most biometric testing is done by developers» Traditional single vendor, proprietary technology testing» Interoperability testing Measurements of interoperability» Uniquely done by independent test labs

8 Biometric Interoperability is Different The Internet: Routers from different companies are interoperable because A standard, IPv4, regulates the packet format Competitive marketplace But also the receiving router doesn t care about content Modulo policies on network non neutrality Biometrics: For example, fingerprint minutiae encoders and matchers from different companies follow A standard, INCITS 378:2004, regulating the format The receiving matcher is sensitive to the content Algorithmic dependency has accuracy implications

9 Interoperability of Minutia Templates False Non Match Rate at False Match Rate of 0.01 Supplier of Verification Template + Template Matcher NEC Supplier of Enrollment Template NEC Red values refer to NATIVE performance : One vendor generates and matches all templates. Figures from MINEX 04:

10 Interoperability of Minutia Templates False Non Match Rate at False Match Rate of 0.01 Supplier of Verification Template + Template Matcher NEC Sagem Supplier of Enrollment Template NEC Sagem Red values refer to NATIVE performance : One vendor generates and matches all templates.

11 Interoperability of Minutia Templates False Non Match Rate at False Match Rate of 0.01 Supplier of Verification Template + Template Matcher NEC Sagem Cogent Supplier of Enrollment Template NEC Sagem Cogent Red values refer to NATIVE performance : One vendor generates both templates and matches them.

12 » MINEX MINEX Definitions» An NIST industry collaboration to support the INCITS 378 and ISO/IEC standards through evaluation driven research, development and standardization» A family of tests aimed at supporting standardized fingerprint minutia templates as the de facto leading biometric element for identity management» Ongoing assessment of the core algorithmic accuracy and interoperability of minutia matching algorithms» Calibration services

13 Interested parties MINEX

14 About NIST Testing

15 NIST Tests and Related Image Group FRVT (face) ICE (iris) 1:N Fingerprint Quality Support Standardization Test Lab Accreditation MBGC X Y MINEX (minutiae) Test Standards IREX (iris) US Gov. Systems PFT End Stage Slap Segmentation ELFT Latent MINEX I (2004) MINEX II (Cards) IREX 08 ( ) X Y ELFT I (Lights out) ELFT II (CDEFFS) Ongoing (PIV) Interop MOC Conform XML 29120

16 MINEX II The NIST Context NIST Support for Compact Biometric Elements e.g. for Identity Credentials sbmoc MINEX IREX Standards SC37 WG3 MINEX I 2004 Initial evaluation Ongoing MINEX PIV MINEX II Match on Card MINEX III Minutia quality calibration

17 NIST s Offline Testing» Massive archival image databases support» Scientific Method Reproducible results Fair comparative evaluations» Statistical significance e.g. in low FMR measurements» Can test FMR 10 4» But» Forces algorithms to deal with standard images: decouples sensor from the recognition algorithm Is this realistic?» Offline image sets don t usually capture the transactional nature But MOC matching is expensive

18 System on card» Sensor is on the card» Testing by human attempts only» Expensive to verify that FMR 10 4 Fidelica id-smart Biometric Associates Scanecotech Fidelica on Divacard

19 Testing MXC vs. MOC vs. SOC Match off card» On PC, on wall reader, on network server» GOOD: Algorithm could be certified in lab process» Augmented MINEX I» BAD: But fielded installation cannot be tested for FMR compliance» Is algorithm the same?» Is threshold the same? Match on card» On card» GOOD: Algorithm could be certified in lab process» Augmented MINEX II» GOOD: Production card stock can be sampled and» Testing for compliance with certified product» Need PI System on card» No external reader» BAD: Algorithm can only be certified in a scenario test» Many attempts with a large human population = expense» O(10 5 ) people to check FMR = 10 4» Or market forces» An operational test?

20 MINEX II

21 MINEX II Why MOC?» Match on Card Why» Cards are ubiquitous» ISO/IEC 7816 cards Cryptographically hardened token FIPS certified» Privacy Enhancing Technology (PET) 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

22 MINEX II Why?» Hypothesis:» Match on Card implementations have same accuracy as Match off Card» Why might that be?» MOC is not new.» Many of the same companies 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 a definitive, public, independent, simultaneous measurement of the algorithmic accuracy and speed of MOC implementations

23 Not in MINEX II Scope Also needed» Card reliability, robustness» Card vulnerability» Security evaluation» Image match on card?» Conformance to 7816 x» Contact vs. contactless And there s alternatives» Proprietary templates» System on card» Business model, economics

24 MINEX II Design objectives» Make it: independent, statistically robust, repeatable» 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 compact card three bytes per minutia

25 MINEX II Card APDUs Reference Template: sent via PUT DATA Verification Template sent via VERIFY FNMR FMR Similarity Score via GET DATA

26 MINEX II Evaluation Principle In vitro» 1. Measure accuracy by Execute N template comparisons on general purpose computer In vivo» 2. Confirm by repeating n «N comparisons on the card, check output is identical n = O(10 3 ) N = O(10 6 )

27 MINEX II Status Phase I Phase II Phase III ISO Standard» July Oct, 07» Nov 07 Feb 08 Δ Nov 08 May 09 Δ August 08» Precise Biometrics + TecSec» Neurotechnology + Internet Risk Mgmt» Sagem Orga» Oberthur + id3» Giesecke Devrient» Sagem Orga» Oberthur + id3 Δ Δ Gemalto + Micro Packs Gemalto + Cogent Δ Gemalto + Innovatrics Δ Oberthur + id3 Δ ISO/IEC CD Performance Testing of 7816 based Algorithms Δ Not fingerprint specific» Non public» Public Report NIST Interagency Report 7477 Δ Public Report May 2009 as NISTIR 7477 (Rev) Δ Not standard template specific Participant = Team of { Card vendor + Fingerprint matcher vendor }

28 genuine impostor VERIFY times (seconds) G+D Oberthur / id3 Gemalto / Cogent SAGEM Orga Oberthur / id3 Gemalto / Innovatrics Gemalto / Micro Packs

29 MINEX II PIV Compliance?» PIV Compliance requires matchers to compare templates from other producers with low recognition error rates.» Row denotes enroller, column denotes matcher.» Green indicates accuracy is better than PIV requirement.

30 MINEX II Results» Protocol» Vendor acceptance Eight teams 19 implementations» Open source support For ISO STD prep» 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» Two implementations attain PIV compliance Four others are close» Median VERIFY execution time < 0.5s» Speed accuracy tradeoff is alive and well, but supplier influence is larger

31 Thank You MINEX Root MINEX II IREX Feedback is welcome:

32 MINEX II

33 MINEX II Data + Conclusions Experimental scope» Accuracy + timing metrics» Comparisons of operational DOS vs. DHS images POE BVA» O(10 5 ) subjects» O(10 5 ) genuine comparisons (i.e. no reuse)» O(10 5 ) impostor comparisons (i.e. no reuse)» Left and right index fingers» Collected with quality in the loop auto capture» Non habituated population Results» Accuracy approaches that of match off card» Some matchers much more accurate than others» Median VERIFY times» 0.54 secs (most accurate)» 0.18 secs (fastest)» False non matches at FMR = are 2 4 times more frequent than at FMR = 0.01.» Second finger is needed for only 4% of genuine users

34 MINEX II Why?» Agencies asked us to (DHS CIO, DHS TSA)» Others interested (DOS)» ISO/IEC 7816 cards» Are ubiquitous» Cryptographically hardened token» FIPS certified» Privacy Enhancing Technology (PET)» No central database» Biometric reference never leaves the card» To be the first to do it» Beating U. Bologna

35 MINEX II Why not?» Verification template must be made off card» And passed to the card» So encryption needed SBMOC» A matcher on every credential» Scarce computational resources» So questions about algorithmic accuracy MINEX II

36 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

37 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 compact card three bytes per minutia

38 Match on Card Issues There is a need to test Limited computational resources Smaller templates MOC uses fewer minutiae Reduced angular resolution in ISO format Guide implementers using MINEX data: Pruning (also relevant to small sensors) Angular quantization Evaluation approach Are comparisons done in vivo (on cards) equal to comparisons done in vitro (on PC, as in MINEX)

39 NIST MOC Testing Match on Card Testing ISO/IEC 7816 (initially) Assurance that matching is being conducted on card Economic issues # Cards in issuance vs. # readers. Logical vs. Physical access. Network infrastructure. Security issues PIN release Key management Certification issues Enrollment template and matcher from same company? Depends on: Proprietary template benefit Is that vendor s template good? Standards activity SC 17 ISO/IEC SC 37. ISO/IEC compact card format header req

40 MINEX II Software support for MOC» MOC Template support» Transcoding INCITS 378 to ISO CC templates:» ISO/IEC 7816 support» MINEX II interface uses (PUT DATA, VERIFY etc)» ISO/IEC {2,5,6} support» 2005» Open source here:

41 MINEX II Software support for MOC MOC Template Support Transcoding INCITS 378 to ISO CC templates: p.html ISO/IEC 7816 Support MINEX II interface uses (PUT DATA, VERIFY etc) See And the open source test driver here l

42 ISO/IEC

43 Revision ISO/IEC :2005 Minutia» New definitions of minutiae placement» Simplified record vs. card vs. normal types» Clarification that three bytes per minutia is only for ISO/IEC 7816 smart cards» New field for sensor ID» CBEFF is no longer normative» Parent images of different sizes

44 MINEX Minutiae Interoperability Exchange Evaluation based support for standards Patrick Grother Management Review, May 8, 2008

45 Minutiae from two detection algorithms Insertion / Deletion Angle Difference Angle, position and type Differences

46 Registered Traveler US Iris camera Card (ISO/IEC 7816) Card reader

47 INCITS 378 as Parents to ISO CC User presents card Reader requests BIT from card Reader prompts for specific finger Scan produce output image Template extraction produces INCITS 378 Remove N K minutiae based on quality + polar distance, per BIT Quantize minutia angle (8 6 bits) Quantize (x,y) pix cm 1 Sort minutiae (XY, YX, Polar), per BIT ISO/IEC compact card template Send to card Match Decision

48 Biometric Interoperability Enrolment sample, collected for visa or passport issuance. Produced by Product A STANDARD DATA Matching by Product C Verification sample collected during border crossing. Produced by Product B STANDARD DATA Decision to admit based on verification decision Here, sample could be an image or a template from any modality When A = B = C the sample can be fully proprietary. Otherwise standards are needed.

49 Three way interoperability Enrolment Template Produced by Dermalog Sagem Matcher Authentication Template Produced by NEC Measurements of interoperability» Most biometric testing is done by developers» Traditional single vendor, proprietary technology testing» Interoperability testing» Uniquely done by independent test labs

50 MINEX Compliance (Eligible for GSA Certification) Minutia Template Generators » 20 companies passed with at least one implementation Minutia Matchers » 17 companies passed with at least one implementation» 8 companies failed with one or more implementations» 12 companies failed with one or more implementations

51 Historical Perspective 1986 ANSI/NBS ICST Minutiae Based Identification 1993 November 2000 June 2000 July 2002 January 2003 October 2004 August 2005 February 2006 April 2007 July ANSI/NIST CSL Image Based Identification AAMVA Driver License/Identification Card ANSI/NIST ITL Establishment of M1 then SC 37 Completion INCITS 378:2004 in M1 The minutia standard 2003 December Request from industry to test INCITS 378:2004 HSPD 12 FIPS 201 PIV 2004 April Publication of ISO/IEC February 2005 April 2006 March MINEX I Final Report SP PIV Biometric Data Specification Publication of ISO/IEC MINEX I Continuing Phase starts 2007 March Request from industry to test Match on Card (TWIC/PIV) MINEX II (MOC) Publication of Test Plan 2007 November Completion of ISO/IEC Testing Standard 2008 February MINEX II (MOC) Final Report Criminal applications Civil applications Demand Driver Demand Driver

52 MINEX I» Initiated in response to request from industry» FBI, DHS, interested: DHS, DOJ funded» Goal:» Assess whether standardized fingerprint minutia templates (0.3KB) could be viable replacements for fingerprint images (8KB)» Viable means Accurate and Interoperable Is the standardized information sufficient? Is the standardized information exchangeable?

53 Conclusions» FNMR is lowest when both templates and the matcher come from the same supplier ( native )» FNMR is lower when both templates come from one supplier» Template generation is idiosyncratic» Syntactic conformance is not enough for interoperability» Some template generators are semantically non conformant» Non conformance is evident in the 2D minutiae occurrence density.» Such non conformance degrades interoperability» Single image template analysis is necessary to explain empirical MINEX results further» Minutiae level feedback is essential for improvement» Offline technology testing is suited to measurement of core algorithmic interoperability

54 So Minutia Interoperability is incomplete. Is there an alternative? Yes Images, in applications where they can be exchanged Broadband networks And no In applications where images are too big, or too slow to process Smartcards particularly match on card Band limited networks

55 Conclusions» Large marketplaces are evolving around standardized biometric data records.» Existing markets largely built on non standardized data.» Demand for (good) standards currently exceeds supply» Biometric standards are a different!» Interoperability depends on the content (cf. IPv4)» Template standards incur an accuracy loss (vs. parent image)» Cross vendor comparison of standardized samples => accuracy loss» Biometric standards have not historically been developed with testing in the loop» The MINEX and IREX tests support development» Testing itself is insufficiently standardized» There are no procedures to accredit testing laboratories» Even vocabulary is not yet standardized. Piecemeal efforts elsewhere.

56 MINEX II It will also address next steps in match on card testing, and relate the MINEX II activities to the ongoing revision of the INCITS 378 and ISO/IEC standards. The talk will also give an update on those standards.

57 Minutiae Interoperability Exchange Test (MINEX) An Evaluation of the INCITS 378 Fingerprint Minutiae Templates Patrick Grother NIST MTIT Meeting Brussels, Friday March 23, 2007

58 Overview of Talk MINEX context MINEX primer MINEX major results Recent results Toward an explanation Recent minutia standardization activities Conclusions Next steps

59 MINEX Overview» INCITS 378 standard templates» MIN:A templates encodes coordinates (x, y), angle (θ), type, (no quality)» MIN:B templates MIN:A data plus ridge count, core, and delta information» Proprietary templates» Individual vendor s representation of images» Performance test» Interoperable performance is stated in terms of verification accuracy (FNMR vs. FMR)» 4 datasets» POEBVA, DHS2, POE, and DOS» Left and right index fingers» Test size» match genuine comparisons» non match impostor comparisons» Multiple vendors» 14 vendors in proprietary testing» 14 vendors in MIN:A testing» 6 participants in MIN:B testing

60 » MINEX is intended MINEX Purpose» To assess performance of the new INCITS 378 standard INTEROPERABILITY to assess core capability of algorithms matching standard templates against those generated by other suppliers implementations Template competence SUFFICIENCY to compare performance of algorithms based on standardized vs. proprietary (i.e. image based) templates» MINEX is not intended» To predict performance of PIV, TWIC, RT Why not? Actual performance is dependent on environment, habituation, multiple attempts

61 MINEX Did Not Use the quality field (one zero byte) Differentiate between the enrollment template and the verification template Compare (proprietary, standard) Allow vendors do execute the fusion Did not allow proprietary templates Mandate ridge count / core / delta data Seen as important for 1:N Iterate with vendors

62 PIV» GSA is using MINEX results for its PIV certification» MINEX Report 2006 defined» Accuracy criteria» Interoperable grouping criteria.» Ongoing MINEX program allows:» New entrants» Elevation of performance

63 Other results» Scenario 1 matcher and one template from same supplier» Fixed threshold, set by native comparison FNMR increases, FMR decreases» Scenario 2 three way interoperability» Performance is degraded over two way» Scenario 3 benefit if a device could include multiple template generators» Matchers tend to prefer two of a kind» Scenario 4 for analysis only» There are observable false rejects when templates are produced from the same image

64 MINEX Interoperability :: Why?» Examine cause and effect» Constraints:» Not particularly useful to seek to explain why particular image or template pairs do not match well» Rather need reasons for systematic failure i.e. observations of why two products do not interoperate, on aggregate.» Seek testable property of a system that either indicates nonconformance with the standard or with a requirement

65 INCITS 378 Conformance Clause 2 Conformance A system conforms to this standard if it satisfies the mandatory requirements herein for extraction of minutiae from a fingerprint image as described in Section 5 and the generation of a minutiae data record as described in Section 6. Extraction of minutiae = semantic conformance Check by referencing original image Generation of record = syntactic conformance Check by parsing and validating

66 INCITS 378 on Placement 2004 :: Minutia Placement on a Ridge Ending The minutia for a ridge ending shall be defined as the point of forking of the medial skeleton of the valley area immediately in front of the ridge ending. If the valley area were thinned down to a single pixel wide skeleton, the point where the three legs intersect is the location of the minutia. In simpler terms, the point where the valley Y s, or (equivalently) where the three legs of the thinned valley area intersect :: Minutia Placement on a Ridge Ending The minutia for a ridge ending shall be defined as the point of forking of the medial skeleton of the valley area immediately in front of the ridge ending. If the valley area were thinned down to a single pixel wide skeleton, the point where the three legs intersect is the location of the minutia. In simpler terms, the point where the valley Y s, or (equivalently) where the three legs of the thinned valley area intersect. A Ridge Ending shall be encoded only if all of the legs used to calculate the minutiae angle length (as defined in Angle of a Ridge Ending) are greater than or equal to 0.02 inches in length. The standards contain analogous text for bifurcations also.

67 Interoperability Models :: Results» Model 1: Adjusted R 2 = 0.49» Model 2: Adjusted R 2 = 0.60» Non uniformity (high frequency content) positively contributes to excess FNMR» Non uniformity in both the enrollment and verification templates negatively contributes to FNMR» Regional overlap negatively contributes to excess FNMR» The matcher significantly contributes to excess FNMR, positively and negatively» All effects above are strongly significant» The regression is imperfect» There are missing explanatory variables (minutiae angle encoding differences and other).

68 Angular Encoding Difference θ Given an empirical estimate of the distribution of RMS θ, regress excess FNMR against some statistic (e.g. 95% percentile).

69 Next Steps NIST is interested in: Improved interoperable accuracy Expanding the marketplace of interoperable products Enhanced certification criteria How to achieve this: Collaboration with industry Non public testing Specialized feedback to vendors Reference data (including templates) Refining the standards Does quality field have a role for improved false / missing minutiae problems Leverage MTIT techniques, guidance.

70 Ways forward, for example Do nothing! Interoperability on (x,y,theta) is what it is, and given the commercial and technical constraints, give up. Allow matchers to tailor their processing given knowledge of the incoming template, via the Encoder ID given in INCITS 378 clause 6.4.4, but missing in ISO/IEC Calibration info needed from MINEX / MTIT - like programs Conduct MTIT-like R&D programs, with template exchange and with quantitative feedback to vendors Refine standards: On the quality field, could the encoder produce false minutiae and then reduce using a standardized minutiae quality algorithm. Can one closed template generator be adopted as a reference? Open-source template generator (standardized)? Needs to be demonstrably good.

71 Fingerprint Certification

72 Future Certification Considerations For biometric matching algorithms the following are undesirable:» Existence of wolves» Elevated FMR for some kinds of samples e.g. poor quality Governmental Interest Δ Lower FMR (as a security objective) Δ Usable FNMR (for user convenience) For standard minutia template generators want» Improved conformance of the minutiae to the parent image» placement accuracy» avoid false minutiae» To avoid elevated FMR for low minutiae count samples

73 Matching algorithm weaknesses #1» Effect of number of minutiae in a template at a threshold set to give FMR = No. of minutiae = min(n 1, N 2 )» If a low count template can be caused or introduced, FMR can be elevated by order of magnitude» A vulnerability of some algorithms

74 FMR concentration in subpopulations 3% of enrolled subjects have PERSONAL FMR greater than 15% CDF When threshold set to achieve 5% FMR, the proportion of the population with PERSONAL FMR less than 5% = 60% Personal FMR

75 IREX

76 Iris Recognition» Iris Recognition has received considerable investment and technologies have improved vis à vis» Increased capture volume» Decreased capture time» Segmentation and matching» The promise:» Iris is a second biometric for ID management» 1.5KB 2KB image (published single vendor result)» Alternative to fingerprints for smart cards + band limited networks» Reduced FTA» Interest from Industry + Government» Requirement:» Interoperable iris data image ISO/IEC

77 IREX Family Ongoing development of standardized, high performance, interoperable, iris images i.e. support of ISO/IEC :20XX IREX Now: IREX 08 IREX I Compact Formats A Future Need IREX II Sensor Interoperability A Future Need IREX III Beyond 640x480

78 Iris Recognition IREX 08» Iris equivalent of MINEX I» Support ISO/IEC Iris Image Format» As data suitable for identity credentials (e.g. PIV Card)» As data for transmission over band limited networks» Objective: Evaluate interoperable exchange of» Cropped rectilinear images» Cropped + eyelid and sclera masked images» Concentric circle polar images» Objective: Establish dedicated iris compression profiles» JPEG 2000 or JPEG Compression» The no compression baseline = largest public test of iris recognition algorithms» Driving evolution of iris image standard ISO/IEC

79 Biometric data in US. Registered Traveler Fingerprint minutiae templates 1 to 4 templates INCITS 378:2004 Face photograph A photograph INCITS 385:2004 Polar iris images From 0 to 2 images ISO/IEC

80 » NIST IREX 08 Iris Recognition» Iris equivalent of MINEX I» Driving evolution of ISO/IEC » Iris is a second biometric for ID management Alternative to fingerprints Standoff acquisition is possible Requested by OMB/DHS/Congress

81 The polar problem Because many irides do not have circular inner or outer boundaries, the choice of origin of the polar coordinate system may be subject to arbitrariness, uncertainty, and error. The consequence of indeterminacy in selection of a polar coordinate origin is deformation and distortion in the data, which is irrecoverable and therefore much more serious than a shift error in rectilinear coordinates. we recommend removal of the polar image formats from J. Daugman UK Contribution to SC 37 Doc. N2124 May 2007

82 Proposal #1 The Unsegmented Polar Format Raw image Coarse segmentation Polar wrapping Compression Storage on smartcard Read for verification Decompression Polar unwrapping Fine segmentation Feature extraction Match against new sample

83 Proposal #2 ROI (Region of Interest) Masking Raw image Coarse segmentation Grey value mask Compression Storage on smartcard Read for verification Decompression Fine segmentation Feature extraction Match against new sample

84 Test driven Standards Development NIST Iris Exchange Test 2008 Evaluate proposed formats Baseline against unconstrained methods Results drive revision of ISO/IEC Fall 2008

85 Template standards» ISO/IEC Iris image format Polar transform

86 MINEX II The NIST Context NIST Support for Compact Standardized Biometric Elements e.g. for Identity Credentials sbmoc MINEX IREX Standards SC37 WG3 MINEX I 2004 Initial evaluation Ongoing MINEX PIV MINEX II Match on Card MINEX III Minutia quality calibration MINEX IV Conformance

87 Zebra biometrics Cattle January 1, 2008 * U.S million head * Canada 13.9 million head [Source: USDA]

88 Relative Performance False Non-Match Rate False Match Rate