摘  要: 特征提取是人脸识别过程中的重要步骤，对识别率高低有着非常重要的影响。本文在对主成分分析（PCA）、Fisher线性鉴别分析（FDA）以及核主成分分析（KPCA）三种特征提取方法的优劣性的理论分析研究与实验比较的基础之上，提出一种新的人脸特征提取方法，即PCA+KPCA+FDA（PKFDA）。实验表明，该方法能够有效地提高识别率，并且也没有耗费很多的时间。因此，该方法具有可行性和优越性。

关键词：PCA; KPCA; FDA

A Modified Feature Extraction Method for Face Recognition
 

XUE Yan, LI Jian-liang, ZHU Xue-fang

( Dept. of Information Management, Institute of Multimedia Information Processing, Nanjing Univ., Nanjing 210093)

(Dept. of Mathematics, , Nanjing Univ. of Sci. & Tech. Nanjing, 210093)

Abstract: As is known, a good pattern recognition mathod depends closely on the effectiveness of feature selection and extraction. Combined with the theories analysis and experiment comparison of  the merits and shortcomings of Principal Component Analysis、Kernel Principal Component Analysis and Fisher Discriminant Analysis , a new face recognition method PKFDA is proposed to extract face features in this paper. Some numerical experiments show that this face recognition method is more better than PCA、KPCA and FDA methods. 

Keywords: Principal Component Analysis; Kernel Principal Component Analysis; Fisher Discriminant Analysis