Abstract
Super-resolution is very popular for research in the image processing during past years. It has been widely used on different digital devices such as smart phones, televisions, webcams and digital cameras. The goal of super resolution is to get a high resolution image from one or more low resolution images. In this thesis, we propose a super resolution method that generates a high-resolution image from a single low resolution image without training images by local linear embedding. We use the input low resolution image itself and its down-sample image as training images. In our algorithm, we find several nearest patches for each input low resolution image patch from its down-sample image and compute the reconstruction weights, and then use the weights to reconstruct high resolution image patches with the help of the input low resolution image. In our framework, we also apply the Shapley value to define the nearest patches. Finally, we do several experiments to show that it is well visual quality and the objective criteria, which is represented as the PSNR value.