Kernel-Based Bayesian Face Recognition

The intrapersonal subspace in Bayesian face recognition algorithm is a successful model to face recognition. In the algorithm, the intrapersonal subspace is described by a linear subspace produced by principal component analysis. In this paper, we propose a new kernel-based Bayesian face recognition algorithm which defines the intrapersonal subspace after a nonlinear map and constructs it by nonlinear component analysis. The ¿kernel trick¿ is used for the algorithm can be expressed by dot product. We prove that the original Bayesian face recognition algorithm is just a special case of the new algorithm. Experiments of the algorithm on the FERET database show an encouraging recognition performance of the new algorithm.