文献翻译指纹识别系统Word文档下载推荐.docx
《文献翻译指纹识别系统Word文档下载推荐.docx》由会员分享,可在线阅读,更多相关《文献翻译指纹识别系统Word文档下载推荐.docx(16页珍藏版)》请在冰豆网上搜索。
histogramequalization;
ridgethinning;
ridgeending;
ridgebifurcation.
1.INTRODUCTION
Fingerprintrecognitionsystemsaretermedundertheumbrellaofbiometrics.Biometricrecognitionreferstothedistinctivephysiological(e.g.fingerprint,face,iris,retina)andbehavioral
(e.g.signature,gait)characteristics,calledbiometricidentifiersorsimplybiometrics,forautomaticallyrecognizingindividuals.In1893,itwasdiscoveredthatnotwoindividualshavesamefingerprints.Afterthisdiscoveryfingerprintswereusedincriminalidentificationandtillnowfingerprintsareextensivelyusedinvariousidentificationapplicationsinvariousfieldsoflife.Fingerprintsaregraphicalflow-likeridgespresentonhumanfingers.Theyarefullyformedataboutseventhmonthoffetusdevelopmentandfingerprintconfigurationdonotchangethroughout
thelifeexceptduetoaccidentssuchasbruisesorcutonfingertips.
Becauseofimmutabilityanduniqueness,theuseoffingerprintsforidentificationhasalwaysbeenofgreatinteresttopatternrecognitionresearchersandlawenforcementagencies.Conventionally,fingerprintrecognitionhasbeenconductedviaeitherstatisticalorsyntacticapproaches.Instatisticalapproachafingerprint’sfeaturesareextractedandstoredinann-dimensionalfeaturevectoranddecisionmakingprocessisdeterminedbysomesimilaritymeasures.Insyntacticapproach,apatternisrepresentedasastring,tree[1],orgraph[2]offingerprintfeaturesorpatternprimitivesandtheirrelations.Thedecisionmakingprocessisthensimplyasyntaxanalysisorparsingprocess.
Thispapersuggeststhestatisticalapproach.Experimentalresultsprovetheeffectivenessofthismethodonacomputerplatform,hencemakingitsuitableforsecurityapplicationswitharelativelysmalldatabase.Thepreprocessingoffingerprintsiscarriedoutusingmodifiedbasicfilteringmethodswhicharesubstantiallygoodenoughforthepurposeofourapplicationswithreasonablecomputationaltime.BlockdiagramforthecompleteprocessisshowninFigure.1.
2.IMAGEPREPROCESSING
Fortheproperandtrueextractionofminutiae,imagequalityisimprovedandimagepreprocessingisnecessaryforthefeaturesextractionbecausewecannotextracttherequiredpointsfromtheoriginalimage.Firstofall,anysortofnoisepresentintheimageisremoved.Orderstatisticsfiltersareusedtoremovethetypeofnoisewhichoccursnormallyatimageacquisition.Afterwardsthefollowingimagepreprocessingtechniquesareappliedtoenhancethefingerprintimagesformatching.
2.1HistogramEqualization
Thismethodisusedwheretheunwantedpartoftheimageismadelighterinintensitysoasto
emphasizethedesiredthedesiredpart.Figure2(a)showstheoriginalimageandFigure2(b)histogramequalizationinwhichthediscontinuitiesinthesmallareasareremoved.Forthehistogramequalization,lettheinputandtheoutputlevelforanarbitrarypixelbeiandl,respectively.Thentheaccumulationofhistogramfrom0toi(0≤i≤255,0≤k≤255)isgivenby
whereH(k)isthenumberofpixelwithgraylevelk,i.e.histogramofanarea,andC(i)isalso
knownascumulativefrequency.
2.2DynamicThresholding
Basicpurposeofthresholdingistoextracttherequiredobjectformthebackground.Thresholdingissimplythemappingofalldatapointshavinggraylevelmorethataveragegraylevel.TheresultsofthresholdingareshowninFigure3.
2.3RidgelineThinning
Beforethefeaturescanbeextracted,thefingerprintshavetobethinnedorskeletonisedsothatallridgesareonepixelthick.Whenapixelisdecidedasaboundarypixel,itisdeleteddirectlyformtheimage[3-5]orflaggedandnotdeleteduntiltheentireimagebeenscanned[6-7].Therearedeficienciesinbothcases.Intheformer,deletionofeachboundarypixelwillchangetheobjectintheimageandhenceaffecttheobjectsymmetrically.Toovercomethisproblem,somethinningalgorithmsuseseveralpassesinonethinningiteration.Eachpassisanoperationtoremoveboundarypixelsfromagivendirection.Pavlidis[8]andFieginandBen-Yosef[9]havedevelopedeffectivealgorithmsusingthismethod.However,boththetimecomplexityandmemoryrequirementwillincrease.Inthelatter,asthepixelsareonlyflagged,thestateofthebitmapattheendofthelastiterationwillusedwhendecidingwhichpixeltodelete.However,ifthisflagmapisnotusedtodecidewhetheracurrentpixelistobedeleted,theinformationgeneratedfromprocessingthepreviouspixelsincurrentiterationwillbelost.Incertainsituationsthefinalskeletonmaybebadlydistorted.Forexample,alinewithtwopixelsmaybecompletelydeleted.Recently,Zhou,QuekandNg[10]haveproposedanalgorithmthatsolvestheproblemdescribedearlierandisfoundtoperformsatisfactorilywhileprovidingareasonablecomputationaltime.ThethinningeffectisillustratedinFigure4
3.FEATURESEXTRACTION
Thetwobasicfeaturesextractedfromtheimageareridgeendingsandridgebifurcation.For
fingerprintimagesusedinautomatedidentification,ridgeendingsandbifurcationarereferredtoasminutiae.Todeterminethelocationofthesefeaturesinthefingerprintimage,a3x3windowmaskisused(Figure5).MisthedetectedpointandX1…X8areitsneighboringpointsinaclockwisedirection.IfXnisablackpixel,thenitsresponseR(n)willbe1orotherwiseitwillbe0.IfMisanending,theresponseofthematrixwillbe
whereR(9)=R
(1).ForMtobeabifurcation,
forexample,ifabifurcationisencounteredduringextraction,maskwillcontainthepixel
informationsuchasR
(1)=R(3)=R(4)=R(6)=R(7)=0,R
(2)=R(5)=R(8)=R(9)=1,and
Foralltheminutiaedetectedintheinterpolatedthinnedimage,thecoordinatesandtheirminutiaetypeissaveasfeaturefile.Attheendoffeatureextraction,afeaturerecordofthefingerprintisformed.
4.MATCHING
Fingerprintmatchingisthecentralpartofthispaper.Theproposedtechniqueisbasedonstructuralmodeloffingerprints[11].Oneofthemajorbreakthroughsofthismethodisitsabilitytomachfingerprintsthatareshifted,rotatedandstretched.Thisisachievedbyadifferentmatchingapproach.Asitisclearthatthisalgorithmmatchesthetwofingerprintimagescapturedatdifferenttime.Thismatchingisbasedontheminutiaeidentificationandminutiaetypematching.Matchingprocedureiscomplexduetotwomainreasons;
1)Theminutiaeofthefingerprintcapturedmayhavedifferentcoordinates
2)Theshapeofthefingerprintcapturedatdifferenttimemaybedifferentduetostretching.
Anautomatedfingerprintidentificationsystemthatisrobustmusthavefollowingcriteria:
1)Sizeoffeaturesfilemustbesmall
2)Algorithmmustbefastandrobust
3)Algorithmmustberotationallyinvariant
4)Algorithmmustberelativelystretchinvariant
Toachievethesecriteria,thestructuralmatchingmethoddescribedbyHrechakandMcHugh[11]isadoptedasthebasisofourrecognitionalgorithm,withchangesmadetothealgorithm,toprovidemorereliableandimprovingoverallmatchingspeed.Thismatchingrepresentsthelocalidentificationapproach,inwhichlocalidentifiedfeatures,theirtypeandorientationissavedinfeaturesfile,iscorrelatedwiththeotherimage’sextractedfeaturesfile.ThemodelisshowninFigure6.
Foreachextractedfeaturesonthefingerprint,aneighborhoodofsomespecifiedradiusRaboutthecentralfeatureisdefinedandthenEuclideandistanceandrelativeanglesbetweenthecentralpointandtheotherpointisnotedwiththepoint’stype.Sincethedistanceamongthepoint
remainsthesamethroughoutthelife.Sothistechniqueworkswellfortherotatedandshiftedimages.
5.CONCLUSION
Afingerprintrecognitionalgorithmthatisfast,accurateandreliablehasbeensuccessfullyimplemented.Thisalgorithmcanbemodified,introducingtheridgelinecount,andthencouldbe
usedinonlineandrealtimeautomatedidentificationandrecognitionsystem.
REFERENCES
[1]MOAYER,B.,andFU,K.S.:
‘Atreesystemapproachforfingerprintpatternrecognition’,IEEETrans.,1986,PAMT-8,(3),pp.376-387
[2]ISENOR,D.K.,andZAKY,S.G.:
‘Fingerprintidentificationusinggraphmatching’,PatternRecognit.,1986,19,
(2)pp.113-122
[3]TAMURA,H.:
‘Acomparisonoflinethinningalgorithmsfromdigitalgeometryviewpoint’.ProceedingsoffourthinternationaljointconferenceonPatternRecognition,Kyoto,Nov.1978,pp.715-719
[4]HILDITCH,C.J.:
’Linearskeletonfromsquarecupboards’,MachineIntel.,1969,4,pp.403-420
[5]NACCACHE,N.J.,andSHINCHAL,R.:
‘AninvestigationintotheskeletonizationapproachofHilditch’,
PatternRecognit.,1984,17,(3),pp.279-284
[6]JANG,B.K.,andCHIN,P.T.:
‘Analysisofthinninga