資源描述：
《基于判別性低秩字典學(xué)習(xí)的稀疏表示人臉圖像識別.pdf》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在教育資源-天天文庫。
1、SparseRepresentationforFaceRecognitionbasedonDiscriminativeLow-RankDictionaryLearningLongMa,ChunhengWang,BaihuaXiao,WenZhouStateKeyLaboratoryofManagementandControlforComplexSystemsInstituteofAutomationChineseAcademyofSciences95ZhongguancunEastRoad,100190,BEIJING,CHINA{long.ma,ch
2、unheng.wang,baihua.xiao,wen.zhou}@ia.ac.cnAbstractmizationproblem:minxs.t.y=Dx(1)1Inthispaper,weproposeadiscriminativelow-rankxdictionarylearningalgorithmforsparserepresentation.whereDisanover-completedictionary,xisthesparseco-Sparserepresentationseeksthesparsestcoef?cientstor
3、ep-ef?cientvector,andyisthetestsignal.Toimprovetheper-resentthetestsignalaslinearcombinationofthebasesinformanceofsparserepresentation,Yang[30]proposedro-anover-completedictionary.Motivatedbylow-rankma-bustsparsecodingtomodelthesparsecodingasasparsity-trixrecoveryandcompletion,a
4、ssumethatthedatafromtheconstrainedrobustregressionproblem;Liu[21]constrainedsamepatternarelinearlycorrelated,ifwestackthesedatathesparsecoef?cientstobenonnegative;Huang[14]ex-pointsascolumnvectorsofadictionary,thenthedictio-ploitedtheclusteringtendsinnonzerocoef?cients.Thesenary
5、shouldbeapproximatelylow-rank.Anobjectivefunc-algorithmsusedtheoff-the-shelfbasesasthedictionary.tionwithsparsecoef?cients,classdiscriminationandrankLearningthedictionaryhasbeenprovedtoimprovethesig-minimizationisproposedandoptimizedduringdictionarynalreconstructiondramatically[
6、8].Severalalgorithmshavelearning.Wehaveappliedthealgorithmforfacerecogni-beenproposedtooptimizetheatoms.Aharon[1]general-tion.Numerousexperimentswithimprovedperformancesizedthek-meansclusteringprocessandproposedK-SVDoverpreviousdictionarylearningmethodsvalidatetheef-algorithm,th
7、ealgorithmiterativelyupdatedthesparsecod-fectivenessoftheproposedalgorithm.ingofthesamplesbasedonthecurrentdictionaryandthenoptimizedthedictionaryatomstobetter?tthedata.Mairal[23]proposedanenergyformulationwithbothsparsere-constructionandclassdiscriminativecomponents.Anon-1.Intr
8、oductionlinedictionarylearning[22]algorithmbase
