采用PCA_ICA特征和SVM分类人脸识别21676.pdfVIP

15 4 V ol15, No4 2003 4 JO U RNAL OF COM PU T ER- AI DED DESIGN COM PU T ER GRA PHI CS Ap , 2003 PCA/ ICA SVM 王宏漫 欧宗瑛 ( CAD CG 116024) , , ; , 1- 1 , VC , , ; ; ; ; T P3914 Face Recognition Based on Features by PCA/ ICA and Classification with SVM Wang H ongman Ou Zongying ( CA D CG L abo ratory , S chool of M echani cal E ng ineer i ng , Dal ian Univer si ty of T echnol ogy , D alian 116024) Abstract T his pape deals w it h com bined app oach of p inciple component analysis ( PCA) and inde- pendent com ponent analysis ( ICA) to get t he ep esentat ion basis of f ace imag e set and p oposes a new mult-i step app oach t o ex tend suppo t vect o machine ( SV M ) capability t o deal w it h mult-i class f ace ecognit ion by inco po ating wit h t he eliminat ion st ategy Based on t he one- ag ainst-one classifying st at eg y, it so ts the disc iminat ion f unct ions acco ding to t hei ow n Vapnik- Che vonenkis confidence and uses t he edundancy among them t o dec ease t he disc imination e o in case of eject ing decision Ex pe iments wit h tw o face-dat abases show t hat t he p oposed met hod has eached a hig he ecognit ion ate w ith easonable t ime cost Key words f ace ecognition; p incipal component analysis; independent component analysis; suppo t vect o machine; elim inat ion st ategy [ 1] 3D 2D 1 , ( ) ( , ) ( ( , ) ) , , , [ 2]