Image Super-Resolution through Pyramid Learning

This paper presents a novel approach to single image super-resolution. We construct two pyramids: low-resolution image pyramid and the corresponding high-resolution image pyramid, then perform image segmentation and cluster the image patches according to a certain rule. We seek a sparse representation for each patch in pyramid via a corresponding dictionary. Our method aims to learn the relationship between the sparse coefficient of low-resolution image patch and that of the corresponding high-resolution image patch using support vector regression (SVR). So the final high-resolution image can be obtained via implementing the learned relationship on the input low-resolution image. Unlike the prior example-based method, our method does not require the external training image data. Also the experiment result display that our method get a better effect than the existing interpolation or example-based method.