Propose an efficient algorithm for low-rank approximation Develop a new method with proved convergence to learn sparsity Incorporate low-rank recovery and sparse coding into one joint framework Validate the proposed algorithm with parameter sensitivity and performance