Efficient Blind Image Super-Resolution

Abstract

A hybrid method to Single Image Super-resolution is proposed. We used zero-shot super-resolution method to reconstruct high-resolution image from low-resolution one based on the degradation trained on unpaired high-resolution and low-resolution samples. This approach gives the benefits of internal networks, such as extracting features from a particular picture, as well as external methods working with high-resolution and low-resolution image distributions. The proposed scheme would be of high-interest for super-resolution of single images from a specific devices with the same degradations.