SVM classifier for face recognition based on unconstrained correlation filter

Abstract

In this paper we present a novel method of face recognition technique using a combination of unconstrained correlation filter and support vector machine. The unconstrained minimum average correlation energy (UMACE) filter generates a recognition parameter based on peak to side lobe ratio (PSR). Instead of training the support vector machine by the face image for classification, the PSR values from a set of UMACE filters is used to train the SVM. The proposed technique is tested with Cropped Yale B illumination database and the method shows significant reduction in error rate compared to classical UMACE filter based technique.