Multi-Scale Multi-Stage Single Image Super-Resolution Reconstruction Algorithm Based on Transformer

In this paper, creatively combining Transformer with image super-resolution reconstruction, we proposes a multi-scale multi-stage single image super-resolution reconstruction algorithm based on Transformer (MSTN). The algorithm uses Transformer as a feature sharing module, thus it realizes network parameter sharing, dynamically focuses on the correlation between feature information of adjacent stages, and then extracts the high-frequency texture information embedded in the current stage features from the feature information learned in the previous stage, which achieves a coarse-to-fine enhancement of image reconstruction. Experiments show that our method can not only per-form better image super-resolution reconstruction compared with other advanced methods, but also reduce the network parameters to a great extent.