Face Recognition Using Kernel Direct Discriminant analysis algorithms

《Face Recognition Using Kernel Direct Discriminant analysis algorithms》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、TOAPPEARINIEEETRANSACTIONSONNEURALNETWORKS,ACCEPTEDINAUGUST2002FaceRecognitionUsingKernelDirectDiscriminantAnalysisAlgorithmsJuweiLu,K.N.Plataniotis,A.N.VenetsanopoulosBellCanadaMultimediaLaboratory,TheEdwardS.RogersSr.DepartmentofElectricalandComputerEngineeringUniversityofTo

2、ronto,Toronto,M5S3G4,ONTARIO,CANADAAugust12,2002DRAFT2SubmittedtotheIEEETransactionsonNeuralNetworksinDecember12,2001.Revisedandre-submittedinJuly16,2002.AcceptedforpublicationinAugust1,2002.PUBLICATIONFORMAT:REGULARPAPERAREA:IMAGEPROCESSINGANDRECOGNITIONCORRESPONDENCEADDRESS:

3、Prof.K.N.PlataniotisBellCanadaMultimediaLaboratoryTheEdwardS.RogersSr.DepartmentofElectricalandComputerEngineeringUniversityofToronto10King’sCollegeRoadToronto,OntarioM5S3G4,CanadaTel:(416)946-5605Fax:(416)978-4425E-mail:kostas@dsp.toronto.eduDRAFTAugust12,20023AbstractTechniq

4、uesthatcanintroducelow-dimensionalfeaturerepresentationwithenhanceddiscriminatorypowerisofparamountimportanceinfacerecognition(FR)systems.Itiswellknownthatthedistributionoffaceimages,underaperceivablevariationinviewpoint,illuminationorfacialexpression,ishighlynonlinearandcompl

5、ex.Itisthereforenotsurprisingthatlineartechniques,suchasthosebasedonPrincipleComponentAnalysis(PCA)orLinearDiscriminantAnalysis(LDA),cannotprovidereliableandrobustsolutionstothoseFRproblemswithcomplexfacevariations.Inthispaper,weproposeakernelmachinebaseddiscriminantanalysisme

6、thod,whichdealswiththenonlinearityofthefacepatterns’distribution.Theproposedmethodalsoeffectivelysolvestheso-called“smallsamplesize”(SSS)problemwhichexistsinmostFRtasks.Thenewalgorithmhasbeentested,intermsofclassificationerrorrateperformance,onthemulti-viewUMISTfacedatabase.Resu

7、ltsindicatethattheproposedmethodologyisabletoachieveexcellentperformancewithonlyaverysmallsetoffeaturesbeingused,anditserrorrateisapproximately34%and48%ofthoseoftwoothercommonlyusedkernelFRapproaches,theKernel-PCA(KPCA)andtheGeneralizedDiscriminantAnalysis(GDA)respectively.Key

8、wordsFaceRecognition(FR),KernelDirectDiscriminantAnalysis(KDD