Kernel based symmetrical principal component analysis for face classification

Kernel method is a powerful technique in machine learning and it has been widely applied to feature extraction and classification. Symmetrical principal component analysis (SPCA) is an excellent feature extraction method for face classification because it utilizes the symmetry of the facial images. This paper presents one Kernel based SPCA (KSPCA) algorithm which gives the closed form for polynomial kernel. KSPCA combines advantages of SPCA with kernel method, i.e., KSPCA not only makes use of the symmetry of the facial images, but also extracts nonlinear principal components which contain more abundant information. We compare the performance of SPCA, kernel PCA (KPCA) with KSPCA on CBCL database for binary classification, and on ORL and Yale face database for multi-category classification, respectively. The experimental results show that KSPCA outperforms both SPCA and KPCA.