Compressive Superresolution Imaging Based on Local and Nonlocal Regularizations

Compressive sensing based on a redundant dictionary has been successfully applied in superresolution imaging. However, due to the neglect of the local and nonlocal interactions of patches of a single image, the reconstructed results are not satisfactory in noise suppression and edge sharpness. In this paper, we propose an improved method by adding steering kernel regression and a nonlocal means filter as two regularization terms and use an efficient clustering subdictionary learning scheme. We further demonstrate better results on true images in terms of traditional image quality assessment metrics.