Dictionary learning for image super-resolution
Juan Li, Jin Wu, Shen Yang, Jin Liu · 2014
Recently, single image super-resolution reconstruction via sparse representation has attracted increasing interest. In this paper, we propose a new method for image super-resolution using a local sparse model on image patches. We introduce a new dictionary training formulation, which enforces that the sparse representation of a low-resolution image patch can well reconstruct its underlying high-resolution image patch, and we adopt an effective stochastic gradient algorithm to solve the corresponding optimization problem. Considering the scale of the recovered high-resolution image patch has been altered in sparse recovery, we introduce an efficient method to find its correct scale. Moreover, the high-resolution deficiency image is reconstructed by the proposed super-resolution method and compensated to better preserve the high-frequency details of images. Compared with the recently proposed joint dictionary learning method for image super-resolution, the experimental results of our method show visual, PSNR and SSIM improvements.