Image super resolution via visual prior based digital image characteristics

Abstract

Designing effective image priors is a key issue to image super-resolution. However, obtaining analytical forms for evaluating the smoothness of the priors is still a difficult and significant task. In this paper, we propose a prior-based method that divides image edges based on the hardness value, replacing the traditional binary classification of edges with a more detailed classification method. Through this partition, we can achieve smoother and better visual effect. Furthermore, we propose a non-uniform refinement approach to effectively improve the speed and reduce the processing time. Experimental results on multiple real world images have demonstrated the advantages of the proposed method over other existing prior methods both in visual effect and processing speed. © 2013 Springer-Verlag.