Facial Recognition System Employing Transform Implementations of Sparse Representation Method

Abstract

A new discriminative sparse representation approach for robust face recognition via l2 regularization (SRFR) was recently published. In this paper, a face recognition system implementation employing coefficients from two non-orthogonal transform domains, namely, Two-Dimensional Discrete Wavelet Transform (2D DWT) and 2D Discrete Cosine Transform (2D DCT), is presented. The use of these coefficients in this Mixed Wavelet Cosine Sparse Representation for Face Recognition (MWCSRFR) system as features shown to appreciably lower the computational complexity and the final storage size while maintaining the high recognition rate of the SRFR. Extensive simulations were carried out on five face databases, namely, ORL, YALE, FERET, Cropped AR, and Georgia Tech. The improved properties of the MWCSRFR are proved as shown in the given sample results.