基于神经网络的血管模式识别分析(IJMSC-V1-N3-2)

I.J.MathematicalSciencesandComputing,2015,3,9-19PublishedOnlineSeptember2015inMECS()DOI:10.5815/ijmsc.2015.03.02Availableonlineat(147002),IndiabProfessor,PunjabiUniversity,Patiala(147002),IndiaAbstractBiometricidentificationusingveinpatternsisarecenttechnique.Theveinpatternsareuniquetoeachindividualevenintwinsandtheydon‟tchangeoverageexcepttheirsize.Asveinsarebeneaththeskinitisdifficulttoforge.BOSPHOROUShandveindatabaseisusedinthiswork.HandveinimagesareuploadedfirstandkeypointsusingScaleInvariantFeatureTransform(SIFT)areextracted.Thentheneuralnetworkisusedfortrainingtheseimages.Finallyneuralnetworkisusedfortestingtheseimagestocheckwhethertheimageusedfortestingmatcheswiththeexistingdatabaseornot.ResultsarecomputedlikeFalseAcceptationRate(FAR),FalseRejectionRate(FRR),accuracyanderrorperbitstream.IndexTerms:NeuralNetwork,ScaleInvariantFeatureTransform,FalseAcceptationRate,FalseRejectionRate,CannyEdgeDetector©2015PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience1.IntroductionBiometricisthetermusedincomputersciencetorefertothefieldofmathematicalanalysisofuniquehumanfeatures[2].Itreferstotheidentificationofhumansbytheircharacteristicsandtraits.TraditionallyauthenticationisbasedontokenbasedidentificationsystemsforexamplePassport,drivinglicenseetc.andknowledgebasedsystemforidentificationsuchasapassword,personalidentificationnumberetc.Asthreatsandattacksareincreasingdaybyday,sothereisaneedofreliablesecuritymechanism.Traditionalmethodsarenotasefficientthatcanprovidesufficientsecurity,sobiometricsystemsarecreatedtoovercomethelimitationsoftraditionalsystems.Biometricsisthescienceofidentifyingapersonusingitsbehavioralandphysiologicalfeatures[2].Biometricssystemsareclassifiedin2categoriesthatarephysicalandbehavioral.Physicalsystemsarerelatedtotheshapeofthebodysuchasfingerprints,facerecognition,DNA,vascularpatterns,irisoftheeyeetc.Behavioralbiometricssystemarerelatedtobehaviorofapersonlikevoice,gaitetc.Advantagesofthesesystemsarethattheyaredifficulttocopyorforge;hencethesesystemsaremoresecure*Correspondingauthor.Tel.:E-mailaddress:10AnalysisofVascularPatternRecognitionUsingNeuralNetworkandoffermorereliableperformance.Vascularpatternisthenetworkofbloodvesselsbeneathaperson‟sskin.Thesevascularpatternscanbeusedtoauthenticatetheidentityofanindividual.Theshapesofvascularpatternsareuniqueineachindividualevenintwins[6].Asbloodvesselsarehiddenbeneaththeskinandarenotvisibletohumaneye,sothesepatternsareveryhardtocopyascomparedtootherbiometrictraitssuchasfingerprints.Vascularpatternscanonlybetakenatlivebody.Allthesecharacteristicsmakethisbiometricsystemmoresecureandreliable.2.PreviousWorkJ.M.Cross,C.L.Smith[1]proposedasystembasedonlowcostautomaticthermographicimagingsystem.Matchingcanbedonebycomparingagivensignatureagainstasingletemplateorlibraryoftemplates.Bysettingtheminimumforwardandreversepercentagesto75%and60%respectively,itconstitutesaFRRof7.5%andFARof0%.T.Tanka[3]proposedasystemusingphaseonlycorrelationandtemplatematchingwhichyieldsFARof5.82%andFRRof16%atthreshold0.40.MohammedShahinetal.[6]proposedasystembasedonspatialcorrelationofhandveinpatterns.TheyobtainedFARof0.02%andFRRof3.00%atthreshold80.ChetnaHegdeetal.[8]proposedasystemtoovercomethelimitationofauthenticationduetodamagedveinsusingmodularizationandconstitutesFARof2.1%andFRRof1.274%.S.Manikandaetal.[10]proposedapalmbasedauthenticationsystemwiththehelpofenergyfeaturebasedonwavelettransform.TheyobtainedFARandFRRbetween0.7to1.0%.V.KrishnaSree,P.SudhakarRao[12]Theyproposedatechniqueinwhichlinearhoughtransformisusedforextractionoffeaturesofqueryanddatabaseimages.FormatchingbetweenqueryimageanddatabaseK-Nearestneighborsearchisused.Theextractionofthesevascularpatternswasobtainedbymorphologicaltechniques.Toenhancetheveinpatternsnoisereductionfiltersareused.InthissystemFARis20%andFRRis3.75%.N.V.Krishnaveniatal.[13]proposedasytembasedonhandveintriangulation.TheyhaveusedBOSPHOROUSdatabaseandobtainedFARbetween1to1.5%andFRRbetween0to1.25%.3.ProposedWorkThehandveinauthenticationapproachinthisworkisbasedontheneuralnetwork.Therearefourstagesinthisauthenticationprocess.(i)ImageAcquisition(ii)ImageNormalization(iii)FeatureExtraction(iv)TrainingofhandveinimagesusingNeuralNetwork(v)TestingwithNeuralNetwork.ThemethodologyoftheproposedworkisshowninFigure1.Fig.1.–MethodologyofproposedworkAnalysisofVascularPatternRecognitionUsingNeuralNetwork113.1ImageAcquisitionImageAcquisitionistheprocessofgettingtheimagesthatarerequiredfortheauthenticationprocess.BOSPHOROUSHandveindatabasewascollectedfromBOGAZICIUniversity,Turkey[4].ThestructureofthehandveinimagescanbecapturedwithtwotypesofimagingtechnologiesnamelyNearInfrared(NIR)imaging,FarInfrared(FIR)imaging[9].FIRtechnologyworkswithintherange8-14nm.Itismoresuitableforcapturinglargeveinsinthebackoft