Deep Unpaired Blind Image Super-Resolution Using Self-supervised Learning and Exemplar Distillation

Existing deep blind image super-resolution (SR) methods usually depend on the paired training data, which is difficult to obtain in real applications. In this paper, we propose an effective unpaired learning method to solve the blind image SR problem. The proposed method first estimates the blur kernel and intermediate high-resolution (HR) image from the low-resolution (LR) input image in a self-supervised learning manner. With the estimated blur kernels and intermediate HR images, we develop an effective variational model based on the image formation of SR to improve the quality of the intermediate HR images. To better learn the LR-to-HR mapping, we further develop an exemplar distillation module that is able to explore useful information from exemplar HR images to constrain the deep models learned from the self-supervised learning module for the final HR image restoration. We jointly train the proposed method and show that it performs favorably against state-of-the-art methods on benchmark datasets and real-world images.