Non-local similarity edge-guided based semi-coupled dictionary learning super resolution

In this paper, we propose a novel edge preserving and noise adaption for retina image superresolution (SR) reconstruction. The proposed method incorporates non-local similarity and edge difference into semicoupled dictionary learning (NSED-SCDL). Firstly, in order to suppress speckle noise during reconstructing, non-local similarity is utilized to construct the denoising constraint. Secondly, for preserving the edge information of the reconstructed image, the edge difference between the observed low-resolution (LR) image and degraded version of the reconstructed image is employed to construct regularization term. Thirdly, we explore the adaptive coefficients of edge constraint to find the optimal edge information during optimizing the objective function. Experiments on retina images demonstrate that the proposed algorithm outperforms other state-of-the-art methods, especially for the noise retina images with weak edges.