Single image de-raining by multi-scale Fourier Transform network

Removing rain streaks from a single image presents a significant challenge due to the spatial variability of the streaks within the rainy image. While data-driven rain removal algorithms have shown promising results, they remain constrained by issues such as heavy reliance on large datasets and limited interpretability. In this paper, we propose a novel approach for single-image de-raining that is guided by Fourier Transform prior knowledge. Our method utilises inherent frequency domain information to efficiently reduce rain streaks and restore image clarity. Initially, the rainy image is decomposed into its amplitude and phase components using the Fourier Transform, where rain streaks predominantly affect the amplitude component. Following this, data-driven algorithms are employed separately to process the amplitude and phase components. Enhanced features are then reconstructed using the inverse Fourier Transform, resulting in improved clarity. Finally, a multi-scale neural network incorporating attention mechanisms at different scales is applied to further refine the processed features, enhancing the robustness of the algorithm. Experimental results demonstrate that our proposed method significantly outperforms existing state-of-the-art approaches, both in qualitative and quantitative evaluations. This innovative strategy effectively combines the strengths of Fourier Transform and data-driven techniques, offering a more interpretable and efficient solution for single-image de-raining (Code: https://github.com/zhengchaobing/DeRain ).