Modular approach on kernel principal component analysis for enhanced face recognition

A novel face recognition approach, modular kernel principal component analysis (MKPCA), combining the idea of modularity in a kernel method is proposed in this paper. In this technique, face images are divided into sub images (modular approach) and features are extracted from a high dimensional space formed using a Gaussian kernel. This method combines advantages of both modular PCA - more local features and kernel PCA - nonlinear modelling of data. Simulation results on standard databases show that the proposed MKPCA method of face recognition out performs PCA, modular PCA and kernel PCA in recognition rates.