Multi-frame Image Super-resolution Reconstruction based on Sparse Representation and POCS

Abstract

Super-resolution image reconstruction algorithms produce a high-resolution image from one or a set of low-resolution images of the desired scene. In this paper, we present a novel two-stage super-resolution (SR) algorithm combined sparse signal representation with the projection onto convex sets (POCS). In the first stage, inspired by recent results in sparse signal representation, we get a high-resolution intermediate image based on learning dictionary method for each low-resolution image of an input image sequence. In the second stage, by fusing these high-resolution intermediate images, a higher resolution image is generated based on POCS method. Experiment results show the effectiveness of our method and the improved performance over other SR algorithms.