Single image super-resolution via convolutional neural network and total variation regularization

Abstract

We propose a super-resolution reconstruction model based on the fusion of convolutional neural networks and regularization constraints. Our model not only takes advantage of the convolutional neural network’s prominent capability for nonlinear mapping between low-resolution and high-resolution images, but also takes the image inherent tendency to have bountiful repeated structural information into accounts. We derive our total variation regularization constraints based on the image local similarity and non-local similarity. Through coalescence of convolutional nerual network and delicately devised adaptive regularization constraints, our model yields a state-of-the-art restoration quality from a single image. Besides, our system can be expanded to tackle more low-level vision problems as well.