Fast Single Image Super Resolution Reconstruction via Image Separation

In this work, a fast single image super resolution reconstruction (SRR) approach via image separation has been proposed. Based on the assumption that the edges, corners, and textures in the image have different mathematical models, we apply different image SRR algorithms to process them individually. Thus, our approach is divided into three steps: 1) separating the given image into cartoon and texture subcomponent by nonlinear filter based image decomposition technique; 2) using improved local-self similarity model based algorithm to interpolate the cartoon subcomponent and the wavelet domain Hidden Markovian Tree (HMT) model based algorithm to zoom the texture subcomponent; and 3) fusing the interpolated cartoon and texture subcomponents together to derive the recovered high-resolution images. Since the decomposition and super resolution algorithms in the proposed approach are mainly based on simple convolution and linear algebra computations, its efficiency can be guaranteed. Also, the simulated and real-life images experiments can validate the performance of our proposed algorithm when compared with some other state-of-the-art super-resolution approaches.