Adaptive Quaternion Network with Contextual Feature Enhancement for Robust Single Image Deraining

Rain streaks can impair object detection and recognition in computer vision. Complex rain patterns challenge current deraining processes, resulting in poor performance. The Adaptive Quaternion Network with Contextual Feature Enhancement (AQN-CFE), a deep learning model, efficiently removes rain from single images. AQN-CFE captures spatial and color features using quaternion representation, and its contextual feature enhancement module preserves fine details in heavy rain images. On the Rain100H dataset, AQN-CFE outperformed previous approaches with a PSNR of 31.8 dB, SSIM of 0.91, and MAE of 3.5. PSNR increases by 2.3 dB, SSIM by 0.04, and MAE decreases by 15%. These results demonstrate that AQN-CFE improves visual clarity, making it appropriate for autonomous driving and video surveillance.