irex + mbe + minex i + minex ii + piv: nist update session
TRANSCRIPT
ChapterI::MBE‐STILL
Large‐ScaleFaceRecogniFonTesFng
MBEOrganizaFon
MBE – STILL-FACE TRACK
MBE‐STILL::What’snewaboutthistest?
• Largestpublic1:Ntestever conducted • N→3million
• EnrolllifeFmehistory • Ofvisaimages(DOS/DHS)
• Ofarrestimages(FBI)
• Vendorexecutesfusion
• ProvideoperaFonalmetadatato thealgorithm • Sex|Height|Weight|Dateof
birth|DateofPicture
• Reportresourceusage: • Time,storage,memory…
• PoseconformancetesFng • QuanJfydeviaJonfromstandard.
• Explicitlysplit1:1tesFng • WithenrolledDB(Jme+aMend)
• WithoutenrolledDB(e‐Passport)
• TestiteraFvely • Providefeedbacktodevelopers
• DriveR&D,improvedperformance
• SupportapplicaFonofstaFsFcal methodsontheenrolleddatabase • “Training”
• FeaturespaceseparaJon, normalizaJon
ChapterI::IREX
TheIrisExchange(IREX)Program SupporFngIRISInteroperability
IREXI::TestedImageFormats
Parent image from camera
KIND 1
Unsegmented polar
Cropped image
KIND 3
Cropped and masked image
KIND 7
Reconstructed rectilinear
KIND 48
KIND 16
Compression+FormatRecommendaFons • Compression–Avoiditwhenyoucan!
• Lossycompressiondoesincrementaldamagetoimages.
• Eithernocompression,orlosslessmaybesufficient.
Lossless for 1:N Apps over networks
RecogniFonErrorUnderCompression FA
LSE
RE
JEC
T R
ATE
Less compression: Lower FRR and FAR
High compression: Higher FRR and FAR
FALSE ACCEPT RATE
In144pagesIREXcovers
• Effectoflossycompression • JPEGvs.JPEG2000
• Limitsoflosslesscompression
• Effectofirisradius
• Howcloselytocroptheiris? • Comparisonofspecializedformats
• Maskedvs.Polar
• Fitforpurpose
• EffectofpupildilaFon • ChangeindilaJon
• Effectofeyelidocclusion • Effectofiris‐pupildisplacement
• Accuracy • ROCs
• Fixedthreshold–effectonFMR andFNMR
• Speedaccuracytradeoffs
• Templatesize
• FalseMatchRateCalibraFon • Howtosetthethreshold
• Effectofdataset
• Algorithminteroperability • EnrollonA–IdenJfyonB
• Imagequalityassessments
• Biometriczoo
IREX–UmbrellaProgramSupporFngIris Interoperability
L1 Pier 2.4 CrossMatch Seek
Can an image acquired on the L1 PIER be matched against an image gathered on another sensor, for example the CrossMatch SEEK?
Image based Interoperability
ChapterII::MINEXI
CoreInteroperabilityofMinuFaTemplates
MinuFaMatching
Minutiae from enrollment image Minutiae from verification image
InteroperabilityofMinuFaTemplates
FalseNon‐MatchRate SupplierofVerificaFonTemplate+ atFalseMatchRate TemplateMatcher
of0.01 NEC
NEC 0.0129
Supplierof Enrollment Template
RedvaluesrefertoNATIVEperformance:Onevendorgeneratesbothtemplates andmatchesthem.
InteroperabilityofMinuFaTemplates
FalseNon‐MatchRate SupplierofVerificaFonTemplate+ atFalseMatchRate TemplateMatcher
of0.01 NEC Sagem
NEC 0.0129 0.0205
Supplierof Enrollment Sagem 0.0316 0.0140 Template
RedvaluesrefertoNATIVEperformance:Onevendorgeneratesbothtemplates andmatchesthem.
InteroperabilityofMinuFaTemplates
FalseNon‐MatchRate SupplierofVerificaFonTemplate+ atFalseMatchRate TemplateMatcher
of0.01 NEC Sagem Cogent
NEC 0.0129 0.0205 0.0300
Supplierof Enrollment Sagem 0.0316 0.0140 0.0207 Template
Cogent 0.0417 0.0225 0.0136
RedvaluesrefertoNATIVEperformance:Onevendorgeneratesbothtemplates andmatchesthem.
MINEX
• FIPS201regulatesPIV • BindingonagenciesunderFISMA
• SupportedbyspecialpublicaFons • SP800‐76‐1 Biometrics
• SP800‐73 PIVCardInterfaces
• MINEXisusedforcerFficaFonagainstfixed performanceinteroperabilityspecificaFon • Of58algorithmssubmiMed
• 32compliantINCITS378templategenerators
• 26compliantINCITS378templatematchers
ChapterIII::MINEXII
On‐cardcomparisonofISO/IEC19794‐2 minuFarecordsonISO/IEC7816smart
cards
MINEXII–AlgorithmsonCards
Verification Template Reference Template: sent via VERIFYsent via PUT DATA
FMR
FNMR DET
Similarity Score via GET DATA
MINEXIIObjecFves
• GivenISO/IEC7816cardsareunpoweredcrypto‐tokens withlimitedcomputaFonalresources • Canon‐cardaccuracyapproachoff‐card?
• What’stheduraJon? • TeststandardizedasISO/IEC19795‐7atfinal(FDIS)ballot.
• Asaby‐product • DefineprocedureforconversionINCITS378asaparentfor ISO/IEC19794‐2:2005compactformat: • DropminuJa→SortminuJa→DropResoluJon→QuanJzetheta
• QuanJfyaccuracyloss
• Opensourcecode
PublicReportdue June2010
MINEXII::Nowinfourthphase
PHASEI Jul07‐Oct,07 11algorithms Precise‐TecSec
Neurotechno‐IRM SagemOrga
Oberthur‐id3
PHASEII Nov07‐Feb08 9algorithms
G+D SagemOrga
Oberthur‐id3
PhaseIII Nov08‐May09 7algorithms Gemalto‐Micro‐Packs
Gemalto‐Cogent Gemalto‐Innovatrix
Oberthur‐id3
PhaseIV Feb19,2010 17algorithms
8cardsuppliers
8algorithm suppliers
Non‐public PublicReport PublicReport May2009as NISTIR7477(Rev)
NISTInteragency Report7477
ParFcipant=TeamofCardvendor+Fingerprintmatcher vendor
ChapterII::PIV
AnupdateontheBiometricComponents ofthePersonalIdenFtyVerificaFon
SpecificaFon
PIV::DocumentaryHistory
• FIPS201 • LegallybindingonUSG
agencies
• PointstoNISTSpecialPubs
• NISTSpecialPub800‐76 • Version‐0:January2006 • Version‐1:February2007 • Version‐2:Spring2010
IrisspecificaFonsforPIV
• SP800‐76‐2specificaFons(underconsideraFon): Cropped• Irisimage(forcard) masked image• Irisimage(forCMS)
• CameraspecificaJons KIND 7
• Performance Cropped • CaptureAPI image
• ConformancetesJng KIND 3
• Commentiswelcome • Now,and • ondraoofSP800‐76‐2
IREXTest(Supportfor1:1and1:N)
• NISTInteragencyReport7629,Sep21,2009 • PerformanceofIrisRecogni0onAlgorithmsonStandardImages
• QuanFtaFvesupportforISO19794‐6standard • Imagesizeisabout3KB(for1:1)and~30KB(for1:N) • Compression,cropping,formarngprofiles • Speed‐accuracytradespace
• TenimplementaFonsofstandardizedinteroperableirisimage format • Num.irisprovidershasexpandx10inlastfiveyears • NumcoretechnologyprovidersinirisexceedsthatforfacerecogniJon
• IrisimageinteroperabilitysuperiorminuFainteroperability • Lessdependencyontheproductthatpreparestherecord
• hpp://iris.nist.gov/irex(orgoogle“irisinteroperability”)
Match‐on‐CardinPIV::NearTerm
• Possiblenear‐termwayforward: • UseaseparateMOCapplicaJonforcardacJvaJon • UseatrimmeddownversionoftheMINEXIIMOCinterface fortheAPDUs • IthasbeenimplementedbynineorganizaJons,andusedsuccessfully • It’sopenlydocumentedinNISTInteragencyReport7485
• UseoftwofingerswouldsaJsfyFIPS140‐2requirementson falsematch
• Contactinterface
• GICScompliant
Match‐on‐CardinPIV::LongerTerm
• Longertermwayforward • FormallyincludeMOCinafutureFIPS201‐2
• ForcardacJvaJon,and • AsanauthenJcaJonmechanism
• LeverageNISTIR7485forinterface • LeverageNISTIR7452forconfidenJality
• OrimplementaJonsof7816securemessaging
• MeetFIPS140‐3(notFIPS140‐2) • ConsiderISO/IEC24787
TheroleofInteroperableImages
The old paradigm
The new paradigm IG 100
AOPTIX
Proprietary template (iris code) typically ≤ 1KB.
Standardized Iris Image Record ISO/IEC 19794-6
2KB to 308KB
Supplier B
Supplier C
Single supplier iris recognition algorithm
Supplier A of iris recognition algorithms
DoD::OperaFonalContext
• QuesFontoTomDeeduringKeynoteQ+A,DODBiometrics+ ForensicsSummit,SanDiego,May12,2009: • “Hasanyprogressbeenmadeongerngthebiometricdatafrom
OCONUStoCONUS?”…“Thisislikemovingabowlingballdowna gardenhose”.[SpecialOps,Ft.Bragg]
• QuesFontobreakoutS&TPanel,May12,2009PM: • “Isanyoneworkingoncompression,reducingthedataforCOMMs”.
[MarineCorpsSys.Command]
Whycompress?
• Smallpipes,datais“big”,manytransacFons • Net‐centric
• DHS::PortsofEntry→IDENT • DoD::BaMlefield→ABIS
• Card‐centric • CredenJals(USG,PIV,FRAC,CAC)::Smartcards,
• Example: • AEGISDestroyertoCONUS.Withachannelat128 kbps,biometric=single640x480uncompressediris ~308KB→19.25seconds
IREXI::Highlights
• 10RecogniFonAlgorithm • Supportinteroperableimages Companies • Replacetemplate‐based
exchange• 9Commercial,1Academic • Standardsconformantproducts • Evaluateformats
ready‐to‐go • Establishlimitsofcompression
• Threeirisdatasets • QuanFtaFvesupportforthe • Approx10000persons,and standards
100000images. • ISOstandardimages
• Largestpublicindependent • BinaryrecordAND
• SemanJcproperJesofimagesevaluaFontodate • ANSI/NIST• Notatestofcamerasor
systems • NotapredicJonofoperaJonal
accuracy
RelevancetoHSPDs
• Existenceofstandardizedinteroperableirisimage recordssupports • CounterterrorismobjecJvesofHSPD6,11,24
• Crossagency/governmentexchange
• QuanFfiedcompressionresponseaddresses • RealoperaJonalbandwidthconstraints(e.g.HSPD‐24) • Useofcompressedirisdatainblue‐forceapplicaJons(e.g.as astandardizeddataelementonPIV/CACcardsforHSPD‐12)
IrisdatatogoonPIVCards
Following the arrangement of fingerprint minutia data on current PIV cards… Two iris in one container.
Tagged biometric container (SP 800-73)
2.65M cards issued 07/2009
CBEFF Header =88 bytes
ISO Iris Image Header ≥107 bytes
ISO Iris Image Data ~2 * 3KB
CBEFF Signature block ~ 500 bytes
ChapterIII::MBE‐STILL
1:NFacetesFngtailoredtotheFBINext GeneraFonIdenFficaFonProgram
Exploitallpriorencounters
Enroll all prior encountersunder a single ID
Identification of a new Image
The MBE-STILL API supports the idea of a MULTIFACE, each tagged with metadata.
Vendor implementation determines fusion strategy.
Transmit to NGI
or ABIS etc
MBE‐2DSTILL::TestDesignObjecFvesI
• Maintainvirtuesoftechnology/offlinetesFng • LevelplayingfieldforcomparaJvetesJng
• Repeatability,traceability
• AddmaturitytothetesFngcapability • Publishare‐usableAPI;Solicitpublic/suppliercommentsonAPI
• MeasureduraJonofallfuncJoncalls
• Measuretemplatesize
• TestiteraJvely::Provideresults+feedbacktovendors
• MakingatechnologytesthaveoperaFonalrealism • UseoperaJonaldata!
• ExtendpopulaJonsizetoN>106
• ExecuteaproperidenJficaJontest • Don’tmodel1:NasbeingN1:1s
• Open‐universe(useimpostors,usetrueimpostors)search
• Makesampledataavailableinadvance
MBE‐2DSTILL::TestDesignObjecFvesII
• TosupportfacerecogniFonaccuracyviadata • ExploitmulJplehistoricalimagesofanindividual
• AllowtheimplementaJontoexecutefusion
• ExploitoperaJonalmetadata::ProvidetoimplementaJon • Dateofcapture|Dateofbirth|Height|Weight|Sex
• AllowimplementaJontoexecutepost‐enrollment processingontheenrolleddatabase • FeaturespacenormalizaJon,forexample.
• TosupportfacerecogniFonaccuracyviastructuringtheAPI toallowalgorithms,SupportPerformance
1:NinFBI(andUSG)
CHARACTERISTICSOFTHEFBITASK • One‐to‐many,withN→108
• AidedbymulFplearrestrecords • i.e.mulJpleenrolledimages
• AfederatedapplicaFon • State+local+others→FBI • Heterogeneousimages
• Aidedbyfacestandards…but • Picturesdon’tconformto
standards • Picturesvarydespitemug‐shot
standards
• Aidedbymetadata • Age,locaJon,gender…
• Herenow,andontheNGIradar
TAILOREDTESTINGTOWARDIT • Runatestinproper1:Nmode
• Est:106people,107images • OperaJonalimages
• RunatestwithK≥1enrollment imagespersubject • Useeventheprofileviews • ImplementaJonexploitssamples
• InformtheSDKastothe properFesoftheimages • Subjectspecific:Ethnicity,sex • Imagespecific:Age,date
• Runfast • PublicreviewofopenAPI • “Thin”,automaJc,testreports • FIFOparJcipaJon
Standards–AnddeviaFonsfrom…
ISO Standard Expression Gaze Too close Pose Angle
Startek Engineering Incorporated Identix Incorporated Integrated Biometrics, Inc. Ultra-Scan Corporation Antheus Technology SAGEM Morpho Inc. Aware, Inc. Precise Biometrics Motorola XTec, Inc. US Biometrics Corporation
Aware, Inc. Secure Design KK VeridtNeurotechnologija
NEC Corporation Cross Match Technologies
MINEXI
• IniFaltestofINCITS378 performance(vs.image) andinteroperability betweenproducts
BIO-key International, Inc. SPEX Forensics SecuGen Corporation Startek Engineering Incorporated 123ID, Inc. Sonda Technologies Ltd. Aware, Inc. SONATEQ Griaule Tecnologia Ltda Precise Biometrics
Startek Engineering Incorporated NITGen Co., Ltd.
01 02 03 04 05 06 07 08 09 10 11
FRVT 2002 FRVT 2006 MBE
M I N E X Ongoing
MINEX II
CORE MINUTIA INTEROP
7816 ON-CARD MATCHING
Agenda
• IrisRecogniFon(TheIREXProgram) • Standardized,tested,mulJ‐vendorimplementedstandard irisimages→expandingmarketplace.
• IrisExchange(IREX)::RelevancetoHSPDs24,12,11,6.
• IBPCConference(NISTMarch1‐5,2010) • Performance: How to define, how to get it, how to test it, howtoprocureit,howit'snotthewholestory.
• FaceRecogniFon(MBE‐STILL) • TesJngtailoredtotheopen‐set1:NidenJficaJonproblem (e.g.criminal/KSTidenJficaJon,frauddetecJon, watchlists).