MRSNet: A Spatial and Channel Attention Integration Network Considering Multi-Resolution Improves Image Denoising

Recently, convolution neural network (CNN) and attention mechanism are widely used in image denoising. A novel CNN feature extraction block is proposed in this paper, namely, MRSBlock, which processes multiple image resolutions and introduces sparse attention. Ablation experiments on MRSBlock show it improves the denoising effect. This paper presents MRSNet based on MRSBlock and channel attention mechanism to achieve an effective combination of sparse attention and channel attention in image denoising. Experiments results show that MRSNet gets competitive peak signal-to-noise ratio (PSNR) results with fewer parameters and computation cost.