Analysis sparsity based single image superresolution

Abstract

The superresolution problem can be formulated as reconstructing a high resolution image from a down-scaled and possibly blurred version. This problem is a highly ill-posed inverse problem. To regularize this ill-posed inverse problem different methods have been used in previous works, where the use of sparse representation has been quite popular recently. Sparse representation for image processing works on the premise that images can be represented as a sparse linear combination of elements from a redundant dictionary. In a pioneering work, dictionary couples which are learned from a set of images have been used to solve the superresolution problem using synthesis sparsity. In this paper we present a new approach to single image superresolution problem by using the analysis sparse representation model. Simulation results indicate that using analysis sparsity model with a learned analysis sparsity operator can be an effective and efficient alternative to the synthesis sparsity for the image superresolution problem.