Single Image Super-Resolution via Classified Sparse Representation

Single image super-resolution reconstruction is a challenging ill-posed inverse problem currently. In this paper, we propose a method based on image classification and sparse representation for single image super-resolution reconstruction. Various images belong to different image category, and each category contains different contents and structures respectively, especially the high-frequency feature. Therefore, we extract the features of the input low-resolution image, and classify it into the corresponding category. Then the high-resolution image is reconstructed by sparse representation with the dictionary trained from the corresponding database, which consists of high-resolution and low-resolution image patch pairs. The experimental results demonstrate that our method achieves better performance in visual effect and qualitative analysis, by comparison with some well-known methods.