Single Image Super-Resolution via Mixed Examples and Sparse Representation

Existing super-resolution (SR) methods can be divided into two classes: the external examples SR and the internal examples SR. Although these two types of methods have been achieved satisfactory results, such methods are limited by their inherent flaws. This paper proposes mixed example selection method for combining the external examples with the internal examples. We cluster the internal examples into K classes, and select the similar external examples for every cluster to enrich the training database. And then we learn K discriminative dictionaries for the K cluster examples. Finally, we reconstruct the low resolution images with the learned discriminative dictionaries. Experiments validate the effectiveness of the proposed method in terms of visual and quantitative assessments.