Weighted Near Neighbor Classes Block-Sparse Representation for Image Recognition

Sparse representation is deeply studied for its robustness and effectiveness, while its complex computation reduces the efficiency, so method combines nearest neighbor, subspace learning with sparse representation to reduce computation. Considering the within-class scatter is small while the intra-class scatter is large in the subspace, and samples’ contributions of the same class are similar, dealing with the problem by class is more feasible. Weighted nearest neighbor classes based block-sparse representation for image recognition is presented in this paper. First selection k neighbor classes and preserve the corresponding coefficients to weight samples of the k classes, then represent the test sample by block-sparse representation for classification in the subspace. The experiments on AR, Yale B, MNIST, PIE database verify the proposed method’s effectiveness and robustness.