An Efficient Reformative Kernel Discriminant Analysis for Face Recognition

An efficient reformative kernel discriminant analysis, namely enhanced kernel discriminant analysis (EKDA), is proposed in this paper. In the proposed algorithm, a novel criterion, i.e., maximizing the class separability both in the feature space and in the projection subspace, is presented to enhance the discriminant power of KDA. EKDA is more adaptive to the input data under the novel criterion compared with KDA, which enhances the performance of EKDA. Experiments conducted on the Yale and ORL face databases give the higher recognition performance compared with KDA.