Deep sparse representation for Super-Resolution Image Reconstruction

Abstract

Image reconstruction is an important research direction in computer vision. In this paper, a deep sparse representation model is proposed for super-resolution image reconstruction. We firstly study the decomposition of sparse coefficients and the construction of over-complete dictionary, and then use the K- VSD algorithm to extract the image sparse feature. Finally the deep feature migration model is designed to refine image features with deep convolutional neural network (CNN). The experiments carry out on the perspective single-channel, multi-channel and pixel-wise amplitude reconstruction. Both subjective assessments and objective metrics demonstrate that the proposed method has a good reconstruction effect.