RSPNet: Rain Streak Prior Based Rain Removal Network

Single image rain removal remains challenging due to the complex nature of rain streaks and their interaction with background textures.This study proposes RSPNet, a novel rain removal network that leverages high-frequency information as a rain streak prior.By incorporating a Butterworth filter-derived guidance map, RSPNet enhances rain streak features while preserving background details.The network architecture includes specialized modules for high-frequency feature enhancement, channel feature enhancement, multi-scale feature extraction, and feature refinement.Experiments on synthetic datasets (Rain100H, Rain100L, Rain200H, Rain200L, Rain12) and real-world rainy images demonstrate RSPNet's superior performance over state-of-the-art methods.Notably, RSPNet achieves the highest PSNR and SSIM scores across all tested datasets, with significant improvements in visual quality and detail preservation.The proposed method also shows enhanced performance in downstream tasks such as object detection and image segmentation on rain-affected images.