Discriminative locality-constrained sparse representation for robust face recognition

Abstract

Abstract In this paper, a new joint sparse representation method called discriminative locality-constrained sparse representation (DLSR) is proposed for robust face recognition. DLSR incorporates locality and label information of training samples into the framework of sparse representation. Locality information can distinguish dissimilarity between samples and plays an important role in image classification. Compared with the existing methods, DLSR contains more discriminative information of samples and can obtain more discriminative recognition results. Due to the use of l2-norm regularization, DLSR can obtain a closed-form solution. This makes it computationally very efficient. Experimental results based on the benchmark face databases ORL have shown that DLSR can achieve more promising performance than some state-of-the-art methods.