New learning-based super resolution utilizing total variation regularization method

In this paper, we propose a new learning-based approach for super resolution image reconstruction utilizing total variation regularization method. By using the total variation (TV) regularization decomposition, we obtain the structure component which consists of edge component and the texture component which does not include edge component of the image. We use the texture component for the learning-based method instead of high frequency component. The experimental results show improved performance, short computational time, and robustness to the noise compared with the conventional learning-based method.