Single Image Super Resolution Based On Sparse Represetation And Non Sub-sampled Contourlet Transform

Image super-resolution reconstruction refers to reconstruct a corresponding high-resolution image from the existing low-resolution images. This technology can improve the image super-resolution to improve visual effect and to prepare for the subsequent image processing only by the image processing method of software without improve hardware conditions of the imaging device. So this technology has broad application prospects in remote sensing needs of the target, military, medical, security monitoring and the fields where need high-resolution images.This paper focuses study on a single image super-resolution reconstruction algorithm, first introduce the history of the super-resolution image reconstruction studies, as well as super-resolution reconstruction of basic principles and key technologies. In view of image super-resolution via learning is the focus of current research, but also image super-resolution methods based on sparse representation are one kind of them, which have been achieved the best effect of super-resolution currently. This paper is based on super-resolution reconstruction algorithm for sparse representation, analysis revealed that effective low-resolution image feature representation is the key part of solving the high-resolution images based on sparse representation image super-resolution. Therefore, improving the image feature representation module in the conventional super-resolution algorithms, introduced the Non Sub-sampled Contourlet Transform(NSCT) method, considering that it has the advantages of multi-scale multi-directional in signal processing, it was applied in image feature representation for this thesis. A novel single image super-resolution algorithm based on sparse representation and NSCT is proposed.Firstly, the low-resolution image is decomposed into many sub-bands by NSCT. Secondly, extracting each sub-band characteristics and combining them into a feature to represent the low-resolution image. Thirdly, input the feature as priori information into the frame which is based on sparse representation super-resolution algorithm to predict and solve the corresponding high-resolution images. Finally, super-resolution effect on the proposed method is compared with the previous methods via experiments and applying the proposed algorithm for medical image pattern recognition to verify its practical effect.