Subclass discriminant nonnegative matrix factorization for facial image analysis

《Subclass discriminant nonnegative matrix factorization for facial image analysis》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、PatternRecognition45(2012)4080–4091ContentslistsavailableatSciVerseScienceDirectPatternRecognitionjournalhomepage:www.elsevier.com/locate/prSubclassdiscriminantNonnegativeMatrixFactorizationforfacialimageanalysisb,abb,ab,a,nSymeonNikitidis,AnastasiosTefas

2、,NikosNikolaidis,IoannisPitasaInformaticsandTelematicsInstitute,CenterforResearchandTechnology,Hellas,GreecebDepartmentofInformatics,AristotleUniversityofThessaloniki,GreecearticleinfoabstractArticlehistory:NonnegativeMatrixFactorization(NMF)isamongthemos

3、tpopularsubspacemethods,widelyusedinReceived4October2011avarietyofimageprocessingproblems.Recently,adiscriminantNMFmethodthatincorporatesLinearReceivedinrevisedformDiscriminantAnalysisinspiredcriteriahasbeenproposed,whichachievesanefficientdecompositionof2

4、1March2012theprovideddatatoitsdiscriminantparts,thusenhancingclassificationperformance.However,thisAccepted26April2012approachpossessescertainlimitations,sinceitassumesthattheunderlyingdatadistributionisAvailableonline16May2012unimodal,whichisoftenunrealis

5、tic.Toremedythislimitation,weregardthatdatainsideeachclassKeywords:haveamultimodaldistribution,thusformingclustersandusecriteriainspiredbyClusteringbasedNonnegativeMatrixFactorizationDiscriminantAnalysis.Theproposedmethodincorporatesappropriatediscriminan

6、tconstraintsintheSubclassdiscriminantanalysisNMFdecompositioncostfunctioninordertoaddresstheproblemoffindingdiscriminantprojectionsMultiplicativeupdatesthatenhanceclassseparabilityinthereduceddimensionalprojectionspace,whiletakingintoaccountFacialexpressio

7、nrecognitionsubclassinformation.ThedevelopedalgorithmhasbeenappliedforbothfacialexpressionandfaceFacerecognitionrecognitiononthreepopulardatabases.Experimentalresultsverifiedthatitsuccessfullyidentifieddiscriminantfacialparts,thusenhancingrecognitionperform

8、ance.&2012ElsevierLtd.Allrightsreserved.1.Introductiontowhichtheyieldedrepresentationissparse.Towardsthisdirection,HoyerincorporatedthenotionofsparsenessintotheNonnegativeMatrixFactorization(NMF)[1],isanunsuper-stan