Image super-resolution based on patches structure

Due to compression sensing wide research, sparse representation was well applied in super-resolution. At present, super-resolution via sparse representation is performed on the all patches of the image, but it is high time-consuming and its results are not perfect sometime. Considering the difference in the structure of the image patches, we proposed a super-resolution based on image patches structure feature. The method classifies the patches into smooth region and non-smooth region based on gray variance of the patch, and high-resolution and low-resolution dictionaries, for super-resolution via sparse representation, are obtained based on patches structure, and then bicubic interpolation and super-resolution via sparse representation are used to smooth region and non-smooth region respectively. Experiments show that this method can obtain better results and the run-time of image super-resolution is reduced comparison with the Yang et al. [9] method.