Image Recognition Using Manifold Constrained Collaborative Representation

Image recognition is still a challenging task due to the existed illumination and view variations. Manifold learning and representation based classifiers (RCs) are two widely utilized methods to treat the image recognition. The common RCs only emphasize the representation by the training samples globally, while the geometric manifold structure of samples is not fully considered. In this letter, a novel manifold constrained collaborative representation is proposed, which aims to make the representation of query sample be similar with the codes of their nearby-points. Thus, the obtained representations can be more discriminative for recognition. Extensive experiments on several popular databases show that the our proposed method is promising in recognizing various images.