Abstract: This paper proposed a novel dictionary learning method for single image super resolution based on sparse representation．We tried to utilize patch-level clustering to enhance the contextual information in atom learning stage．Unlike the previous dictionary learning works using the image classification,our training set is constructed from the high-resolution and low-resolution patch pairs labeled by different patch-level class,which is more appropriate for image reconstruction．This approach tried to promote the transfer ability of the dictionary which is built on a limited training set and can eliminate the atoms redundancy introduced by multiple training subsets.

Key words: Single image super resolution,Sparse representation,Contextual dictionary,Patch-level clustering