Super-resolution Image Restoration Based on Nonlocal Sparse Coding

Super-resolution image restoration methods based on Compressive Sensing (CS) generally adopt local sparse coding strategy. Such strategy encodes each image block independently, which easily induces artificial blocking effect. To overcome this problem, a super-resolution image restoration method based on nonlocal sparse coding is proposed. The nonlocal self-similarity of image is considered as a prior in the dictionary training and image coding processes, respectively. Specifically, the proposed algorithm trains the dictionary with interpolated low-resolution images, and calculates the weighted average local code of similar patches, in order to obtain the nonlocal sparse code of each image block. Numerical experiments suggest that the proposed algorithm has a good recovery performance, and is robust to image noise.