一种新颖的基于LDA的人脸识别方法.pdf

《一种新颖的基于LDA的人脸识别方法.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、第22卷第5期红外与毫米波学报Voi.22，No.52003年l0月J.InfraredMiiiim.WavesOctober，2003ANOVELFACERECOGNITIONMETHODBASEDONLINEARDISCRIMINANTANALYSISZHANGYan-kunLIUChong-oing（InstituteofImageProcessingandPatternRecognition，ShanghaiJiaotongUniversity，Shanghai200030，China）AbstractAnoveifacerecognitionapproa

2、chbasedontheDCTandtheLDAisproposed.FirsttheDCTisusedfordimen-sionreduction，thenLDAtransformisperformedontheiowerspacetoextractfeature.TheORLfacedatabaseandtheSJ-TU-IPPRfacedatabaseareusedtotestourapproachandthecorrectrecognitionratesof97.5%and92.6%areobtainedrespectiveiy，whichshowstha

3、tourapproachiscomparabiewithotherapproaches.Keywordsfacerecognition，iineardiscriminantanaiysis，discretecosinetransform，principaicomponentanaiysis.一种新颖的基于LDA的人脸识别方法张燕昆刘重庆（上海交通大学图像处理与模式识别研究所，上海，200030）摘要提出一种基于离散余弦变换（DCT）与LDA相结合的人脸识别方法，首先利用DCT将图像进行降维，然后在低维空间中利用LDA进行特征提取.利用ORL人脸数据库和我们上海交通

4、大学图像处理与模式识别研究所的人脸数据库进行测试，实验结果分别得到了97.5%和92.6%的正确识别率，表明它可以和其他方法相比较.关键词人脸识别，线性判别分析，离散余弦变换，主元分析.tures.ButthefirstPCAsteppotentiaiiyiosesusefuiIntroductiondiscriminantinformationthatisimportantforthefoi-Facerecognitionhasbeenofgreatinterestinre-iowingLDA［2］.AnotherdrawbackofthePCAisitscen

5、tyearsbecauseitswiderangeofappiicationsuchascomputationaicompiexity［3］.surveiiianceandsecurity，identityauthenticationandToovercomethedrawbacksofthetraditionaiPCAaccesscontroi，etc［l］.Numerousmethodshavebeen+LDAapproach，weproposeanoveiLDAaigorithmproposedforfacerecognitionintheiastdecad

6、e.Amongforfacerecognitioninthispaper.Thenewaigorithmu-aiitheseapproaches，atechniguebasedoniineardis-sesthediscretecosinetransformation（DCT）torepiacecriminantanaiysis（LDA），whichwasaisocaiiedthePCAforthedimensionreductionatthefirststep，fisherfaces，hasbeenprovedtobeapromisingap-andthenpe

7、rformtheLDAtomaximizethediscriminant［2］proach.power.WeconducttheexperimentonORLdatabaseHowever，duetothehighdimensionaiityimageandSJTUfacedatabase.Theresuitsshowthatourap-spaceandthenumberofimagesinthetrainingsetisproachiseffective.muchsmaiierthanthenumberofpixeisineachim-1LDAforFaceRe

8、cogni