关键词: 核主成分分析, KL距离, 类间距离, 类内差异

Abstract: In order to solve the problem that Kernel Principal Component Analysis face recognition algorithm is sensitive to the change of the overall features, a KPCA face recognition algorithm based on KL divergence is proposed. Using KL distance between-classes distance and within-class dissimilarities are defined. A nonlinear optimization function is established, through that between-classes distance can be maximized, while within-class dissimilarities are minimized, and extracted features are more compact and separable, which is applied to KPCA algorithm. Finally via using ORL face image database the performance of the algorithm is tested. The experimental results show that the algorithm compared with traditional KPCA algorithm has better recognition effect and stability.

Key words: Kernel Principal Component Analysis（KPCA）, KL divergence, between-classes discrimination, within-class dissimilarities