A vector based approximation of KLT and its application to face recognition

Abstract

A face recognition scheme is proposed in this paper. The method utilizes a vector based approximation of KLT (VKLT) which eliminates the large memory demands and the singularity of the covariance matrix matters that are the main drawbacks of the "eigenface method" (a face recognition scheme based on KL transform). The reconstruction error of VKLT approaches the one of KLT preserving also its data dependence, which is important for discriminating face images from non-face images. Moreover, the greater advantage of VKLT over KLT is that of keeping intra-class variations small and consequently increasing the robustness of the face recognition system.