Single Image Deraining Network Based on Non-local Wide-area Region Blocks

This paper presents a single image deraining network based on non-local wide-area region blocks, which integrates a Dual Attention Complementary Enhancement Block (DACEB) for multi-scale feature fusion and Wide Rectangular Region Blocks (WRRB) for capturing long-range dependencies. The proposed method effectively addresses the limitations of existing approaches in nonlinear rain streak modeling and detail preservation, demonstrating significant performance improvements in complex scenarios such as heavy rain and dense rain conditions. Experimental results show that our method achieves PSNR of $\mathbf{3 0. 4 3 ~ d B}$ and SSIM of $\mathbf{0. 8 9 6}$ on Rain100H and Rain100L datasets, outperforming current supervised methods while eliminating the need for iterative optimization and meeting realtime processing requirements.