Image Super-resolution Reconstruction based on Deep Learning and Sparse Representation

This paper addresses the problem of super-resolution(SR) image reconstruction based on sparse representation and deep learning. we approached this problem from the dictionary learning. Firstly, in order to realize the correspondence between the sparse representation coefficients, we proposed the method of joint dictionary learning based on Sparse Denoising Auto-Encoders(NSDAE). Secondly, at the stage of reconstruction, in order to achieve high frequency compensation, we proposed the algorithm of iterative error back projection. Finally, experimental results show that the recovered high-resolution image is competitive in quality to images produced by other SR methods.