In this paper, we study the basic theory of adaptive-normalized convolution and the defect of this method. A novel video sequence super-resolution reconstruction algorithm based on nonlocal normalized convolution is proposed to solve the edge blurred and artifacts in SR reconstruction result. This algorithm can be combined with gray-value information and structural details in the original image. The density of sampled data and local structure in edge decide the shape and size of neighborhood in a pixel, so as to design certainty function and structural-adaptive applicability function. Experimental results prove that our proposed algorithm can provide improved denoising effect in the output result and achieve a state-of-art optical resolution in the image edges and detailed features.
展开▼
