A Novel Approach to Improve the Collaborative Representation for Face Recognition

Abstract

Collaborative representation based classification (CRC) codes a test sample as a linear combination of all the training samples. However, the recognition rate of CRC is not ideal when available training samples are insufficient, as the test sample cannot be accurately represented by limited training samples. In order to address this issue, we propose a novel idea of training samples fragmentation. First, all training samples are divided into two new training sample sets according to the similarity among them. Next, the test sample simultaneously uses the two new training sample sets to perform CRC, which ultimately uses 2M “nearest neighbors” from the two training sample sets to represent and classify the test sample. In addition, this method also takes advantages of a new fusion classification mechanism based on histogram similarity and Euclidean distance, which has been proven to perform better than Euclidean distance classification. The experimental results reveal that the proposed method performs better in face recognition compared with the most representation based classification methods.