Learning a discriminative dictionary for locality constrained coding and sparse representation

Motivated by image reconstruction, sparse representation based classification (SRC) and locality-constrained linear coding (LLC) have been shown to be effective methods for applications. In this paper, we propose a new dictionary learning and sparse representation approach. During sparse coding step, we incorporate locality on representation samples, which preserves local data structure, resulting in improved classification. In dictionary learning step, a `discriminative´ sparse coding error criterion and an `optimal´ classification performance criterion are added into the objective function for better discriminating power. Experimental results demonstrate that our algorithm outperforms many recently proposed sparse representation techniques for face and SAR recognition.