Super-resolution from one single low-resolution image based on R-KSVD and example-based algorithm

Abstract

With the development of sparse coding and compressive sensing, image Super-resolution (SR) reconstruction attracts extensive attentions. In this paper, we mainly focus on recovering super-resolution version given only one single low-resolution (LR) image. The proposed method is combined with the example-based algorithm, which also exploits the relationship between the low image patches and the high image patches. Firstly, the proposed method applies guided filter, the first-order and second-order derivatives to extract multiple features from LR images, which superior to using only one feature space. Then, the effective dictionary is constructed by a novel algorithm called Relaxation K-SVD (R-KSVD). R-KSVD relaxes the constraints of Orthogonal Matching Pursuit method (R-OMP) in training dictionary for K-SVD algorithm. Finally, a new approach is presented to estimating better HR residual image in the Back Projection. Experimental results demonstrate the superiority of our algorithm in both visual fidelity and numerical measures. © 2013 Springer-Verlag.