Sparse representation and collaborative representation have been widely used in face recognition (FR). Collaborative Representation based Classification (CRC) is superior to Sparse Representation based Classification (SRC) in both accuracy and complexity. It is the collaborative representation (CR) mechanism rather than l_1-minimization improves recognition rate in FR. In this paper, based on K-nearest neighbor (KNN), we find K most similar images as the projective subspace for testing sample. Then we propose a new algorithm named Locally Collaborative Representation based Classification in Similar Subspace (LCRC_SS), which changes the projective space from global space to local similarity subspace. The main advantages lie in LCRC SS are making full use of "similar" resources and discarding the redundant "dissimilar" images in CR. Extensive experiments show that LCRC SS has better recognition rate than CRC.
展开▼
