Image super-resolution by exploring spatial and frequency information

Image super-resolution aims to reconstruct high-resolution images from low-resolution images, which is an significant task in image processing. Existing mainstream image super-resolution methods often consider spatial information while ignoring frequency information, which cannot capture high-frequency structural information in images. To tackle this problem, in this paper, we propose a novel image super-resolution method SFSR by exploring both the spatial and frequency information. To achieve this, we design a spatial frequency feature extraction module, which applies the U-shaped network to capture the multi-scale spatial information and utilizes the discrete Fourier Transform based ViT to extract the frequency information. Then, we design a residual fusion module, which can fully integrate the representations obtained from the spatial domain and frequency domain, to obtain more informative representation for super-resolution. At last, we apply a decoder to reconstruct the high-resolution images. We proposed experiments on several benchmark data sets by comparing it with some mainstream image super-resolution methods, demonstrating the suggested method's superiority and efficacy.