On the Use of Monogenic Scale Space for Efficient Face Representation and Recognition

Abstract

In this paper, we present a novel monogenic scale space based Principal Component Analysis (PCA) method by integrating the Reisz transform of face images at different scales and the PCA method for face representation and recognition. The Reisz transform captures desirable facial features characterized by local phase information and local energy at different scales in order to cope with the variations due to illumination and facial pose changes. The PCA method is then employed to reduce the dimension of the feature vectors and hence for efficient face representation and recognition. The feasibility of the proposed monogenic scale space based method integrated with PCA has been successfully tested on many standard face databases such as AT&T and YALE face databases. The recognition accuracy of the proposed approach is compared with the other well known face recognition approaches namely the PCA method, the kernel PCA method and the Gabor wavelet-based RCM method and it is found that the proposed approach exhibit better recognition accuracy when compared to these well known methods.