Adaptive robust sparse representation for face recognition based on weighted and fusion dictionary

We propose a new model for face recognition under insufficient sampling conditions in this paper. In the proposed method, we combine the fusion dictionary with nuclear norm regularization to preserve the details of the restored images, and adopt a Laplacian-uniform mixture function to fit the error distribution. Since the proposed model is convex and separable, we employ the classic alternating direction method of multipliers to solve it by introducing auxiliary variables to transform the original problem into the saddle point problem. Theoretically, we conduct the convergence analysis of the proposed numerical algorithm. Final experimental comparisons are provided to verify the satisfactory performance of the proposed model, which outperforms other related competitive methods in both recognition rate and the robustness.