Image super-resolution via adaptive sparse representation

Jianwei Zhao, Heping Hu, Feilong Cao

Abstract: Existing methods for image super-resolution (SR) usually use ℓ1-regularization and ℓ2-regularization to emphasize the sparsity and the correlation, respectively. In order to coordinate the sparsity and correlation synthetically, this paper proposes an adaptive sparse coding based super-resolution method, named ASCSR method, by means of establishing a regularization model, which effectively integrates sparsity and correlation as a regularization term in the model, and adaptively harmonizes the sparse representation and the collaborative representation. The method can balance the relation between the sparsity and collaboration adaptively via producing a suitable coefficient. To approximate the optimal solution of the model, we adopt a current popular and effective method, i.e., the alternating direction method of multipliers (ADMM). Compared with some other existing SR methods, the experimental results demonstrate that the proposed ASCSR method possesses outstanding performance in term of reconstruction effect, stability to the dictionary, and the noise immunity.