Graph regularized dictionary for single image super-resolution

Super-resolution (SR) for single image is wild used in image processing areas. The learning-based methods use the co-trained dictionaries which contain low resolution and corresponding high resolution images to conduct SR. In this paper, a new dictionary for SR is proposed which adds the graph information between patches. Simulation results show that our scheme improved the dictionary and outperforms the existing classic SR algorithms in both subjective visually and quantitative evaluations.