Super-Resolution Employing an Efficient Nonlocal Prior

Abstract

In this paper, we propose a novel approach for multiframe super-resolution reconstruction by incorporating non-local prior in the maximum a posteriori (MAP) formulation. This prior expresses that recovered images tend to exhibit repetitive structures. A great deal of computation is required in the original non-local prior algorithm dealing with the huge amount of weight calculations. Techniques of weight symmetry, moving averaging filter, limited search window are adopted to speed up non-local filter. Meanwhile, Non-Linear Conjugated Gradient (NLCG) method is introduced to solve simultaneously the high-resolution (HR) image of optimization process and non-local prior adapted to the HR image. Experimental results on extensive synthetic and realistic images demonstrate the superiority of the proposed algorithm to representative algorithms both quantitatively and qualitatively.