Sparse representation for video super-resolution

Sparse representations of signals have been developed rapidly in recent years, such as in the field of super-resolution. The theory that an image is able to be decomposed into a sparse representation over an over-complete dictionary of atoms ensures the feasibility of such applications. In this paper, we produce a pair of well-trained dictionaries that has both high and low resolution patches followed by restoration using sparse representations, and then the steering kernel regression is added into the framework to restrict the whole video and reduce noise and block effects. The experiments show that the quality of recovered super-resolution video is better acceptable and competitive.