DeterminationofMinutiaeScoresforFingerprintImageApplicationsP. Bhowmi - PDF document

DeterminationofMinutiaeScoresforFingerprintImageApplications�P. Bhowmick. B. Bhattacharya
,M.K.Kundu,C.A.Murthypartha t,bishnu t,bhargab,malay,murthy@isical.ac.inIndianStatisticalInstitute,203B.T.Road,Calcutta-700108andT.Acharyatinku acharya@ieee.orgElutionTechnologies,Phoenix,AZ-85226,USAAbstractMany ThisworkisfundedbyagrantfromIntelCorp.,USA(PO#CAC042717000)\bAuthorforcorrespondence.[11],and,forusingdirectionalimagesbyMehtreetal[10].InaworkbyHung[5],ridgeenhancementisdonebasedonridgedirections,andnoiseremovalandpatternpurica-tionareperformedwiththehelp Camera / SensorImage BinarySkeleton MinutiaeExtraction ScoreEvaluationMatchingResultReferenceFingerprintDatabaseFigure1:GenericstructureofanAFISwithminutiaescores.minutiaeapropostheirqualityeitherinthedatasetorinthequeryset.Amatchisreportedifthecoordinates,typesandanglesofminutiaeofqueryset\narefoundtobeagreeingwiththoseofdataset undercertaintransformationsliketranslation,rotation,orscaling[4,6,7,8,12].Inordertoconsidertherelativequalityofaminutiainangerprintimageasapracticalmatchingcriterion,wedeneaminutiapointasa5-tuple,
 \n\r,where,()=coordinatesof,=typeofminutia(abifurcationminutiaoraterminationminutiaasconsideredbytheFederalBureauofInvestigationandadoptedinmostAFIS),\n=anglemadebythetangenttothecorrespondingridgeatthepoint(),and\r=anintegerscoreassociatedwiththeminutia.Thescorevaluesarenormalizedwithinascaleof1to100,where,aminutiawithscorenearing100isofthehigh-estsignicancecomparedtoanyotherminutiawithalowerscorevalue.Inotherwords,ifaminutiahasascore\r,andanotherminutiahasascore\r,where\r\r,thenisalessdependableminutiathan.Whileapplyingamatchingprocedurebasedonnger-printminutiae,thescoresofminutiaeof andthoseof\ncanbeusedtotellabouthowgoodorbadthematchis.Ifaminutia(   )withscore\r inset isapotentialmatchwithaminutia( !"#)withscore\r#inset\n,thedifference\r$ &%'\r#indicatesthequalityofmatchingof( ")and(# #).Foramatchingbetween and\nwith(minutiae,(*),+,wedenethematchingindex-asfollows:-.
0/21143/(5687:9;9�=?A@687:B;B�=CADE\r$ F3G\r$#ESince/IHJ\rKHL/21M1and/NHO\rKHL/21M1for/IHPQHR(,so1SHE\rF3G\rEH,/211,andtherefore,-alsoliesintherange[0,100].Ahighvalueof-impliesastrongmatchbetween and\n,whereas,alowvalueindicatesapoorone.Theconceptofscorecanbealsoexploitedtoexpeditethematchingprocedurebetweenaqueryset\nandadataset .Theproblemistocheckforamatchingin ,ifatallexists,inthengerprintimagedatabase,withrespecttothequeryset\n.Inthatcase,asmallsubset\nofminutiaewithleadingscorevaluesinthequeryset\nshouldbeconsideredrsttocheckforamatchwiththedataset .Ifthematchbetween and\nTissatisfactory,anextlevelmatchcanbetriedbetween andalargersubsetof\n.Thismaybecontinuedtillthereisatotalmatchbetween and\n.Atanyintermediatematchingstageinvolving,say, and\nU,ifthematchisnotsatisfactory,theremainingsetofminutiae,i.e.,\nV3\nTneednotbetriedfor,thussavingthematchingtimeforanunsuccessfulcase.Ascore-basedgenericstructureofanAFISisshowninFig.1.3.EvaluationofscoreThescore\rofaminutiaisestimatedbasedonthefol-lowingproperties:-patternofridgeowinandaround;-patternofvalleyowinandaround;-noiselevelinthelocalityof.Iftheridgeandvalleylinesinthelocalneighbourhoodofhaveasmoothnatureofow,thecorrespondingminutiawillhaveagenuinecontributioninthengerprintmatching.Onthecontrary,ifinsomeregion,theridgeandvalleylineshaveanerraticorunevennatureofow,aminutiaXWinthatregionshouldnotpredominatethematchingprocedure.Theformerminutia(),beinglocatedinatidyregion,con-tributesmorecondenceinthematchingprocedurethanthelatter(IW)whichislocatedinanoisyregion.ForaminutiaZY[,thescoreisgivenbytheequation\r4
O\r\:]_^*\r`"a^b\r2cd(1)where,\r\],\r`"and\rcMdarethescorecomponentsduetoridgeow,valleyow,andnoiselevelrespectivelyinthelocalneighborhoodof.Thecomponents\r\:]and\r`"denotemeasuresofperfectnessofridgeandvalleyowre-spectively,thatareevaluatedbasedonsomedistancesesti-matedinthelocalridgeandvalleytopographyaroundtheminutia.Totakeintoaccountthenoiseoftheregioninandaround,thecomponent\rcMdisestimatedinalocalwindowcenteredat.Noiseimpartsanegativeeffectonthescore.3.1.ScoreofabifurcationminutiaLetebetheaverageinter-ridgedistanceofangerprintim-age.First,wendthethreeneighborpixelsf,f,fXgof,considering8-neighborhood.f,f,fXgarethethreestartingpixelsoftheridgesh,h,hgrespectively,incidentat.Weexploreawalkalongeachofh,h,hgstartingfromfS,fN,fgrespectively,eachwalkbeingoflengthe. Letthesewalksbenamedas,a,andgrespectively.Ifduringsomewalk]A/ZHLP4H,+,anybifurcationorter-minationminutiaisencountered,thewalkishalted.Let,
]A/HPUH+,denotethelengthofthewalk].Let,
]cbetheminimumof
]A/HPH+,andbethenumberofwalkswhoselengthsarelessthane.Ifisaminutiaofgoodquality,theneach
]shouldbeatleaste,andatleasttwoofthemshouldbee.So,if
]ce \nor,O)2,weassign0to\r\r$handreturnfromthispoint.Else,if
]ce,thenwewalkforalength
]calongeachofthethreeridgesh,h2,hgstartingfromfZ,fN,fgre-spectively,sothatafterthe(re-)walks,eachofthepointsg,reachedonthethreeridgesh,h,hgrespec-tively,isatequaldistancefrom(Fig.2).r2r1Q=Q3PQ1Q2r=r3d12d31d23Figure2:Ridgesincidentatthebifurcationminutia.Inthisscenario,weneedtoidentifytheridgelinethatbifurcatesat.InFig.2,thethreeridgesareshownash,h2,andh,where,w.l.g.,h(=hg)hasbeendepictedasthepre-bifurcatedridge,andh,h2areitstwobifurca-tionsat.Toidentifythepre-bifurcatedridge,wedene]c
&P(YMgg2[,where,]=-distancebe-tween]and,/RH'PH'+P
.Ifandareonthetwobifurcatedridgeshandh,thengandg.However,thisconditionmayfailifisapoorminutiacandidate,viz.,whentheridgesinci-dentatareofunevennature,anditisdifculttoascer-tainthepre-bifurcatedridgeamongh,h,h2g.Hence,if]c+
]c!\n,weassign0to\r"#\r$h$,andreturn.qPP'P"QK1L1L2K2n1r1r2vv1v2n2ryxM1V1R1M2V2R2Q2Q1S1S2Figure3:Ridgeandvalleycharacteristicsaroundabifurca-tionminutia.Inordertocomputethescore\r\]forabifurcationminu-tia,wedenethefollowingdistances,videFig.3.&%c'=distancefromtoneighborridge(=)(;%c\n*=distancefromtoneighborridge(=)(;%'c'=distancefromtoneighborridge(=-;%*:c\n*=distancefromtoneighborridge(=-;%'\+*=distancefromtobifurcatedridgeh=, ;%*:\'=distancefromtobifurcatedridgeh=",;Foragoodminutia,theabovedistancesshouldbeclosetoe.So,\r\:]isassignedtodependingontheclosenessof-/.10\:]12435%c&'5%c*5%'c'&%*c*&%'\*&%*\6'7w.r.t.e.Thus,forabifurcationminutia,thescorew.r.t.theridgecharacteristicscanbechosenas:\r\:]
98\:]5:D;=Y!e3Ee3-/.10\:]E[&#x;000;?(2)where,8\:]istheridgescoremultiplierforbifurcationminutiae.Similarly,thescore\r$`"forthebifurcationminutiaisbasedonthefollowingsetofdistances.%`'=distancefromtoneighborvalley@=;&%`*=distancefromtoneighborvalley@=;$ABAC=distancefromtovalleyterminationminutiaXW,ifany,lyingnearinbetweenhandh=TW;$ACC\D'=distancefromIWWtobifurcatedridgeh=TWWFEU;$ACC\*=distancefromIWWtobifurcatedridgeh$=TWWFE;$ACC`?'=distancefromIWWtoneighborvalley@=TWWFG;$ACC`*=distancefromIWWtoneighborvalley@=TWWFG_;where,TWWisthepointalongthevalley@atadistanceefromTW,or,abifurcationorterminationof@appearingwithinthetargetwalk-lengthofe.Whiletheparameter3-H.I0\]7representssomekindofinter-ridgedistance,wedeneotherdistancemeasureswithasubtledifference.Distancesintheset3-/.10`"7=3\nACC`B'$ACC`*7areinter-valleydistances,whichshouldbeideallyclosetoe.Theotherset3-H.I0`"7=35%`?'&%`*M\nADAC\nA#CC\6'$ACC\*7containsdistancesfromaridgepointtoavalleyline,orfromavalleypointtoaridgeline,andtherefore,requiresaexibilityintheircontribu-tionto\r`".Hence,distancesintheset3-/.10`"7areverymuchsimilarto3-/.10\:]7asfarastheestimationof\r`"isconcerned.Theircontributiontoscoremaybechosenas:\r`"
J8`"5:'D;K9Y#e3MLLe3-/.10`"LL[&#x;000;#(3)And,thatdueto3-H.I0`"7is\r`"
5:B*D;K9\r:B*D;K9(4) where,\r:B*D;K9ischosenas:
*D;K9 \n\rif\n\n! "#\n%$'&if"\n)( \n \r$"#\n&if"\n)*\r(5)and8F`"isthevalleyscoremultiplierforabifurcationminutia.3.2.ScoreofaterminationminutiaLetbeaterminationminutiaandfbetheadjacentridgepixelof,considering8-neighborhood.Sinceisater-minationminutia,therewillbeonlyoneridgeline,sayh,incidentat[Fig.4].Wewalkalonghstartingfromf,foralengthe,anddesignatethewalkas.Let
denotethelengthofthewalk.Sinceaskeletonizedngerprintim-ageshouldbedevoidofspursandbridges,
shouldalwaysbeequaltoe.LetbethepointontheridgehreachedqPP'P"QK1L1L2K2n1vn2ryxN1N2v1v2Figure4:Ridgeandvalleycharacteristicsaroundatermi-nationminutia.afterthewalk.Forestimationofthescore\r\:]fortheterminationminutiawithrespecttoridgelinesintheregioncontaining,wedenetheset3+.10\:]7offollowingdis-tances.&%c'=distancefromtoneighborridge(=(;%c\n*=distancefromtoneighborridge(=(;Fortobeaterminationminutiaofgoodquality,theabovedistances,shouldbeclosetoe.Thesedistancesarebasicallyinter-ridgedistancessimilarto3-H.10\:]7inthecaseofbifurcationminutiae.Hence,thescore\r\:]isas-signedtobasedonthefollowingequationthatresembleswithEqn.2inform:\r\:]
-,\:]5:/.;;=Y!e3EeZ30+.10\:]E[&#x;000;?(6)where,,_\:]istheridgescoremultiplierforterminationminutiae.Similarly,thescore\r$`"fortheterminationminutiaisbasedontheset30+ .10`"7offollowingdistances.&%`B'=distancefromtoneighborvalley@=Q;&%`*=distancefromtoneighborvalley@=;$ABAC=distancefromtovalleyterminationminutiaXW,ifany,lyingnearinbetween(and(=TTW;ACCc'=distancefromIWWtoneighborridge(=IWWf;ACCc*=distancefromWWtoneighborridge(=WWf;where,TWWisthepointalongthevalley@atadistanceefromTW,or,abifurcationorterminationof@appearingwithinthetargetwalk-lengthofe.Theabovesetofdistancesaremeasuredeitherfromaridgepointtoavalleylineorfromavalleypointtoaridgeline.Hence,theircontributiontoscore\r`"isgivenby:\r`"
5:.;;K9\r:/.;;K9(7)where,\r:.;;K9ischosenas:
.;;K91#\n'if243\n51#\n0 3\n$'&if3\n(1#\n0 $3\n&if3\n*(8)and,_`"isthevalleyscoremultiplierforaterminationminutia.3.3.EstimationofnoiseLetbeabifurcationorterminationminutiahavingapos-itivescoreaftertheevaluationof\r\:]and\r`".Ifdoesnothaveapositivescore,weneednotevaluate\rcMd,since\rcMdwillcontributeanegativescoreto;nallywewillcon-sideronlythesetofminutiaewithpostivescores.Consideracircularwindow6ofradiusE
VfearoundZY[,videFig.5.Let3]E]lieswithin6&#x;000;PK
/D87787 #97bethesetofpoints,witheachpoint]satisfyinganyoneofthefollowing3properties(Fig.5):(i)]isaridgeminutiawith\r\]+\r`"=0;(ii)]isanon-minutiaridgepointhavingthreeormoreridgesincidentuponit;(iii)]iseitheravalleybifurcationoravalleyterminationminutia.PQ1Q8Q4Q7Q1Q12Q6Q10Q2Q11Q9Q3Q5radius=RFigure5:Contributingpoints3$7877A7inanoisywindow6centeredaroundtheminutia.TheabovedenitionenablesustouseE3]7E=9asameasureofnoiselevelinthewindow6centeredaround.Wedeneanotherparameter:,calledthenoisefactor, whichisusedtondthenoisethresholdcd]
givenintheequationbelow,thatwillindicatewhetherornotawindow6associatedwithaminutiaisnoisy:cMd]

-:f(9)If9ishigherthancMd]
in6correspondingto,thenoiselevelin6isconsideredhighenoughandeachpoint],PS
/MB7877"9,isaccountedonebyonefortheirin-dividualcontributiontothenoise-induced(negative)score\rcdof.Thus,Eqn.10canbeusedtond\r]cMdattributedbyeach],andEqn.11sumsuptheindividualscorestocomputethetotalscoreduetonoise.\r]cMd
YEO3YK][[(10)\rcMd
1if9&H 2cMd]
\n]\r\r]cMdif9 2cMd]
(11)where,isthenoisescoremultiplier.InEqn.10,-distancebetweentwopointsY[andY[isgivenby:YY "["2Y_M[[
3Y3 [^Y_3MA[7(12)4.Experimentsandresults5010015020025030035040045050050100150200250300350400450Figure6:AsamplengerprintimagefromNIST14sdb.WeusedseveralngerprintimagesfromtheNISTSpe-cialDatabase14[1]andNISTSpecialDatabase4[13].Inordertokeeptheminutiaescoresoftheorderof100priortonormalization,thevalueof8\:](=8F`")hasbeenchosenas1.00.Forevaluatingthescoreofabifurcationminutia,weneedtocompute6distancesintheset35-/.10\:]7,mea-suredw.r.t.differentridgelines,and7distancesintheset3-H.10\:]7,measuredw.r.t.differentvalleylines.Fornd-ingthescoreofaterminationminutia,weneed2and3suchdistances,inthesets3+ .10\]7and3+ .10`"7,respectively.Thus,inordertohaveparityinthescorevaluesofbifurca-tionandterminationminutiae,wechoose,\]=(6/2)8\:]=3.0and,`"=(7/3)8`"=2.33.Table1:ScoreValuesofMinutiaesl.no.xyTypeAngleScore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ntheestimationofnoise-basedscore,:isacontrollingparamaterthatdecidestheeffectofnoiseonthescore.FromEqn.9,itisevidentthatahighervalueof:willenforcealesserimpactofnoiseinthescore.Onthebasisofourexperimentalresults,wehaveempericallychosenf=2,:=3,and,=/4:=0.33.InFig.6,asamplengerprintimageofsize1/"isshown.ThecorrespondingternaryskeletonimageisshowninFig.7,wherethedarkerlinesrepresenttheridgesandthefaintlinesarevalleys.TheminutiaehavingpositivescoresareshowninFig.7,withthedarknessofaminutiabeingproportionaltoitsscore.Table1includesthescores(posi-tivevaluesonly)ofthebifurcationminutiae(BM),followedbythoseoftheterminationminutiae(TM),arrangedinas-cendingorders.Thebifurcationminutiaat(297,197)hasthemaximumscore100,whichiswelljustiedbyitsvisualclarityintheimageshowninFig.6andthetopographi-calorderlinessinitsneighborhoodinFig.7.Ontheotherhand,theminutiaat(441,379)islocatedinahighlynoise-affectedregion.ScoresofsomeminutiaearewrittenbesidethecorrespondingminutiaeinFig.7.TheproposedmethodisimplementedinConaSunUltra510,Sparc,+M+-,theOSbeingtheSunOSRelease5.7Generic.ThetotalCPUtimefortheevaluationofscoresofallminutiaeinaternaryskeletonizedngerprintimagewasfoundtobearound0.03to0.07sec. �----- Y-AXIS -----5010015020025030035040045050050100150200250300350400450100 1 90 55 33 66 35 Figure7:Minutiaeshownwithdarknessproportionaltoscores.5.ConclusionsandfutureworksAmethodofscalingtoassessaminutiaforngerprintmatchingisreportedinthispaper.Developmentofafasterandrealisticngerprintmatchingtechniquebasedontheproposedmethodiscurrentlyinprogress.Someoftheem-piricalformulaementionedinthispapermayrequirefurtherrenementsformoreaccuratematchingresult.Inreality,thescoreofaminutiainaqueryimagemaybedrasticallydifferentfromthatofthedatabaseimage.Ifthescoresvarywidely,thenthecondenceinmatchingmayreducesignif-icantly.Theseanomalieshavetoberesolvedtoensureamatchingresult.References[1]G.T.Candela,P.J.Grother,C.I.Watson,R.A.Wilkinson,andC.L.Wilson.PCASYS-APattern-LevelClassica-tionAutomationSystemforFingerprints,NISTIR5647.Na-tionalInstituteofStandardsandTechnology,August1995.[2]A.Farina,Zs.M.Kov´acs-Vajna,andA.Leone.FingerprintMinutiaeExtractionfromskeletonizedbinaryimages.Pat-ternRecognition,vol.32,pages877-889,1999.[3]R.Haralick.RidgesandValleysonDigitalImages.Comput.Vis.Graph.Imag.Process.,vol.22,pages28-38,1983.[4]A.K.HrechakandJ.McHugh.AutomatedFingerprintRecognitionUsingStructuralMatching.PatternRecogni-tion,vol.23,pages893-904,1990.[5]D.C.D.Hung.EnhancementandFeaturePuricationofFin-gerprintImages.PatternRecognition,vol.26,pages1661-1671,1993.[6]A.Jain,L.Hong,andR.Bolle.On-LineFingerprintVeri-cation.IEEETransactionsonPatternAnalysisandMachineIntelligence,vol.19,pages302-313,1997.[7]Zs.M.Kov´acs-Vajna.AFingerprintVericationSystemBasedonTriangularMatchingandDynamicTimeWarping.IEEETransactionsonPatternAnalysisandMachineIntelli-gence,vol.22,pages1266-1276,2000.[8]D.MaioandD.Maltoni.DirectGray-ScaleMinutiaeDetec-tionInFingerprints.IEEETransactionsonPatternAnalysisandMachineIntelligence,vol.19,pages27-39,1997.[9]B.M.MehtreandN.N.Murthy.AMinutiaBasedFinger-printIdenticationSystem.inProceedingsSecondInterna-tionalConferenceonAdvancesinPatternRecognitionandDigitalTechniques,Calcutta1986.[10]B.M.Mehtre,N.N.Murthy,S.Kapoor,andB.Chatterjee.SegmentationofFingerprintImagesUsingDirectionalIm-age.PatternRecognition,vol.20,pages429-435,1987.[11]L.O'GormanandJ.V.Nickerson.AnApproachtoFinger-printFilterDesign.PatternRecognition,vol.22,pages29-38,1989.[12]F.Pernus,S.Kovacic,andL.Gyergyek.Minutiae-BasedFin-gerprintRecognition.inProc.FifthInternationalConferenceonPatternRecognition,pages1380-1382,1980.[13]C.I.WatsonandC.L.Wilson.FingerprintDatabase.NationalInstituteofStandardsandTechnology.SpecialDatabase4,FPDB,April,1992.