Discriminative Sparse Model and Dictionary Learning for Object Category Recognition

Recent researches have well established that sparse signal models have led to outstanding performances in signal, image and video processing tasks. This success is mainly due to the fact that natural signals such as images admit sparse representations of some redundant basis, also called dictionary. This paper focuses on learning discriminative dictionaries instead of reconstructive ones. It has been shown that discriminative dictionaries, which are composed of sparse reconstruction and class discrimination terms, outperform reconstructive ones for image classification tasks. Experimental results in image classification tasks using examples from the Caltech 101 Object Categories show that the proposed method is efficient and can achieve a higher recognition rate than reconstructive methods. Keywords—Sparse representation, dictionary learning, object category recognition, classification