Learning KPCA for face recognition

Kernel principal component analysis (KPCA) is an effective method for face recognition. However, the expression of its final solution needs to take advantage of all training examples, such that its run in real-world application with large scale training set is time-consuming. This paper proposes to apply radial basis function neural network (RBFNN) to learn the feature extraction process of KPCA in order to improve the running efficiency of KPCA-based face recognition system. Experimental results based on two different face benchmark data sets, including ORL and UMIST, show that the proposed method can approach to the recognition accuracy of the original KPCA, but have sparser solutions. The proposed method can be applied to real-time or online face recognition systems.