Abstract
A fundamental challenge in the practical face recognition system lies in determining what, facial features are important, for the identification of faces to overcome the variations in pose, illumination and expression. In this paper, we analyze and extend the traditional DCV to two stages of KDCV (KPCA+DCV) with a novel view namely Kernel Space Isomorphic Mapping, and then apply KDCV to face recognition together with Gabor wavelet to present Common Gabor Vector (CGV) method. Since Kernel method and Gabor analysis are effective to extract the nonlinear facial features, the CGV method is feasible to improve the recognition performance of face recognition. The feasibility of the CGV method is tested on ORL and Yale face databases, and the proposed CGV method gives the highest recognition accuracy compared with the popular PCA, KPCA, LDA, KDA and DCV. Experimental results show that the CGV method is robust to the variations in pose, illlumiation and expression for face recognition.