Super-resolution through non-linear enhancement filters

We proposed a super-resolution system that combines a total variation (TV) filter, shock filter, and learning-based method. In this paper, we ask “what super-resolution should consist of for the texture component of an image?” and propose a new solution that utilizes a TV filter for image decomposition, an improved shock filter for structure component enhancement, and a new non-linear pulse-sharpening filter for texture component enhancement. We obtain good results in terms of picture quality and computational time. We consider this system to be a practical solution, especially for the super-resolution of video signal for devices such as HDTV receivers and PCs with 4K display panels.