Combining sparse representation and local rank constraint for single image super resolution

Sparse representation based reconstruction methods are efficient for single image super resolution. They generally consist of the code stage and the linear combination stage. However, the simple linear combination has not considered the image edge constraint information of image, and hence the classical sparse representation based methods reconstruct the image with the unwanted edge artifacts and the unsharp edges. In this paper, considering that the local rank is able to extract better edge information than other edge operator, we propose a new single image super resolution method by combining the sparse representation and the local rank constraint information. In our method, we first learn the local rank of the HR image via the traditional sparse representation model, and then use it as the edge constraint to restrict the image edges during the linear combination stage to reconstruct the HR image. Furthermore, we propose a nonlocal and global optimization model to further improve the HR image quality. Compared with many state-of-art methods, extensive experimental results validate that the proposed method can obtain the less edge artifacts and sharper edges.