欢迎来到天天文库
浏览记录
ID:37345457
大小:5.00 MB
页数:69页
时间:2019-05-22
《基于PCA和LDA融合的人脸鉴别方法研究》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、华东交通大学硕士学位论文基于PCA和LDA融合的人脸鉴别方法研究姓名:张艳君申请学位级别:硕士专业:计算机应用技术指导教师:宋京伟20060609AbstractSTUDY0FFACEVERIFICATIONMETHODBASEDONFUSlONOFPCAANDLDAABSTRACTFacerecognitiontechnologymakesuseofcomputertoanalysisfaceimages.Itextractsusefulinformationfromfaceimagestodiscriminateidentity.On
2、etaskoffacerecognitiontechnologyisfaceverification.Faceverificationisaprocessofone-to-onematching,itsfunctionistoverifysomebody’Sidentityistrueornot.Inreality,thesecasesexistinmanyfields.Forexample,inadulteducation,whenstudentsgraduate,wehavetoverifytheiridentityusinghisp
3、hotographs.Thesephotographswerecollectedfromentranceexaminationandgraduation,andtheirtimeintervalisaboutfourtofiveyears.Consideringtheworkloadofmanualverifyingareverygreat,theverifyingmethodusingfaceimagesisanurgentdemand.SoitCanusecomputerreplacemanpowertospeedupverifica
4、tion.Inthispaper,itadoptsakindoffacenormalizationmethodwhichcombinesimage’SgreyintegralprojectioncurvesandHoughtransform.Aftertestingbyexperiments,thismethodhassomecharacteristics,suchasrapidspeed,highaccuracyandlessinfluencedbyexpression,deflexionandillumination.Ithas94.
5、9%accuracy.Atthesametime,faceimageshaveshifting,rotatinganddimensioninflexibilityafternormalization.Facenormalizationprovidesanadvantagedguaranteeforfollow-upfeatureextractionandverification.OnthebaseofanalysisandstudyaboutfacerecognitionmethodsbasedonPCAandLDA,weuseaface
6、verificationmethodbasedonthefusionofPCAandLDA.Inthismethod,itintroducesthreefusionmodes.ThereareMax,MaxMinandMeanmode.Thismethodcombinescorrelationanddistinctionofthetwoalgorithmspreferably.Itseriestheproblemoflackingtrainingsetswhichcombinesoriginalfaceimages,its1-orderp
7、rojection·combinedimagesandits2-orderprojection-combinedimages.Bytesting,theverificationresultsofthreefusionmodesinfusionmethodareprovedtooutperformtheindividualverificationalgorithmbasedonPCAorLDA.Itachieves6.39'o^-8.9%higheraccuracythanstandardPCAtechnologyand3.70/o--6.
8、3%higheraccuracythanstandardLDAtechnology.Moreover,itsfalserejectrateis1.05%'-'1.89%lowerthanlow
此文档下载收益归作者所有