采用PCA_ICA特征和SVM分类人脸识别.pdfVIP

第 15 卷 第 4 期 计算机辅助设计与图形学学报 Vol15 , No 4 2003 年 4 月 J OU RNAL OF COMPU TERA ID ED D ESI GN COMPU TER GRA PHICS Apr , 2003 　 采用 PCA/ ICA 特征和 SVM 分类的人脸识别 王宏漫 　　　欧宗瑛 (大连理工大学机械学院 CAD CG 研究所 　大连 　116024) 摘 　要 　人脸识别过程中 ,首先在主成分分析基础上进一步做独立成分分析 ,来提取更加有利于分类的面部特征的 主要独立成分 ;然后采用一种分阶段淘汰的支持向量机分类机制进行识别 该方法扩展了支持向量机处理多类问题 的能力 ,它基于 11 差别策略 ,根据各判别函数 VC 置信范围的差异进行排序 , 同时利用判别函数间的冗余来降低识 别误差 对两组人脸图像库的测试结果表明 ,文中方法在识别率和识别时间等方面都取得了较好的效果 关键词 　人脸识别 ;主成分分析 ;独立成分分析 ;支持向量机 ;淘汰法 中图法分类号 　TP39 14 Face Recognition Based on Features by PCA/ ICA and Classif ication with SVM Wang Hongman 　Ou Zongying ( CA D CG L aboratory , S chool of M echanical Engi neeri ng , D alian University of Technology , D alian 　116024) Abstract 　　This p aper deals wit h combined app roach of p rinciple component analysis ( PCA) and inde pendent component analysis ( ICA) to get t he rep resent ation basis of face image set and p ropo ses a new multistep app roach to extend support vector machine ( SV M) cap abilit y to deal wit h multiclass face recognition by incorporating wit h t he elimination st rategy Based on t he oneagainstone classifying st rategy , it sort s t he discrimination functions accor ding to t heir own Vap nikChervonenkis confidence and uses t he redundancy among t hem to decrease t he discrimination error in case of rej ecting decision Experiment s wit h two facedat abases show t hat t he p ropo sed met hod has reached a higher recognition rate wit h reasonable time co st Key words 　face recognition ; p rincip al component analysis ; independent component analysis ; support vector machine ; elimination st rategy