Image Super-Resolution based on multi-pairs of dictionaries via Patch Prior Guided Clustering

Image super-resolution based on learning dictionary has recently attracted enormous interests. The learning-based methods usually train a pair of dictionaries from low-resolution and high-resolution image patches, ignoring the fact that patches have different structures. In this paper, we propose to train a set of novel multi-pairs of dictionaries for different categories of patches which clustered by gaussian mixture model, instead of a global dictionary trained from all patches. The multi-pairs of dictionaries via patch prior guided clustering can express structure information of the image patches well. Extensive experimental results prove it has strong robustness in super resolution. Compared with state-of-the-art SR methods, our method demonstrates more pleasant quality of image edge structures and texture.