Two-stage mean vector-based competitive representation for image classification

Abstract

Collaborative-competitive representation based classification (CCRC), as a very novel collaborative representation classification method, has achieved good results in the field of image classification. In this paper, in order to enhance the discriminative ability of each class to achieve the improvement of classification performance, we propose a novel collaborative representation (CR) model called the two-stage mean vector-based competitive representation (TMCR) for image classification. The TMCR model introduces a class-specific mean vector constraint and uses a two-stage idea to further improve the classification performance of the model. TMCR divides representation into two stages: coarse representation and fine representation. In the first stage, we select classes in which training samples are similar to the query samples through the mean vector-based competitive representation, and then we use the classes to perform a more correct representation and classification of the query samples in the second stage. In addition, we make full use of the localities of data and introduce weight constraints by calculating the distances between the testing sample and each training sample. The discriminative ability of the model is further improved by the mean vector constraint, the idea of two-stage and weight constraint. Experimental results obtained on GT and ORL datasets indicate that the proposed model can surpass state-of-the-art CR-based methods.