Single Image Superresolution Based On Image Analogy

The goal of image super-resolution is to estimate a high-resolution image from one or more low-resolution images of the same scene (with similar information but different details). It can generate high quality images since it can surmount the reconstruction constraints in imaging process. It has been widely used in remote sensing, high-definition television and security surveillance etc. And it is one of hot research points in image processing and computer vision.In this thesis, we propose a single image super-resolution algorithm based on image analogy and the assumption of self-similarity of image edges and local structures within the same scale and across different scales. This algorithm can be grouped into two frameworks according to the ways of constructing images’pair.First we give a brief introduction of image super-resolution and the research background. Then the advantages and limitations of some state-of-the-art image super-resolution algorithms are analyzed, and some quantitative measures for evaluating the performance of super-resolution algorithms are presented. Based on the assumptions of self-similarity and transfer of local structures within the same scale and across different scales, the image analogy technique can be used to transfer local structures of the input low resolution image to an output high resolution image. In the implementation, we construct image pairs which corresponding to low-resolution and high-resolution versions from the single input image by two different ways, and the estimated mappings between these image pairs can be applied to transfer the input low resolution image to high resolution image. Experimental results show that this super-resolution algorithm can preserve edges, but has distinct ringing artifacts as well, which may be caused by the miss alignment of local structures during transferring, and this will be one of our future works.