A dual sparse and low rank representation for single image super-resolution: A self-learning approach

Abstract

Recently, the sparse representations are one of the most active research areas. Here, the problem of single image super-resolution is revisited with sparse and low rank priors. The introduced algorithm employs a self-learning approach. This self-learning approach is applied on cluster domain rather than the common used patch domain. For supporting the self-learning approach, the learning model adopts an incoherence property with the classical sparse priors. In addition, to compensate the weakness of the high frequency details of the underlying low-resolution image, an edge preserving low lark model is proposed. Hence, the low rank representation guarantees the global structure constraints in the recovered high-resolution images. Experimental results, on different datasets, show that the proposed algorithm can recover high-resolution images compared with the state-of-the art.