Alternating Direction Projections onto Convex Sets for Super-Resolution Image Reconstruction

Abstract

Image reconstruction is important in computer vision and many technologies have been presented to achieve better results. In this paper, gradient information is introduced to define new convex sets. A novel POCS-based model is proposed for super resolution reconstruction. The projection on the convex sets is alternative according to the gray value field and the gradient field. Then the local noise estimation is introduced to determine the threshold adaptively. The efficiency of our proposed model is verified by several numerical experiments. Experimental results show that, the PSNR and the SSIM can be both significantly improved by the proposed model.