W-MTHN: Wavelet-Driven Mamba-Transformer Hybrid Network for Image Deraining

Le résumé fourni par la source

Image deraining plays a vital role in ensuring clear vision under adverse weather conditions. In this paper, we propose W-MTHN, a novel Wavelet-driven Mamba-Transformer Hybrid Network, designed for robust and efficient image deraining. Our key contribution lies in the synergistic integration of a wavelet-based preprocessing stage, which effectively enhances and extracts fine-scale rain patterns, with an advanced hierarchical Transformer-Mamba architecture. Specifically, our model employs a hybrid $\mathbf{2 M L P}+\mathbf{2 M a m b a}$ block configuration in its encoder, along with a direction-adaptive scanning mechanism in its Spatial-Mamba modules, further enhanced by dynamic cross-attention for optimal feature fusion. This unique design enables simultaneous local detail preservation and global dependency modeling, achieving superior deraining performance. Extensive experiments on Raindrop and Rain100H datasets demonstrate that W-MTHN outperforms existing methods, achieving state-of-the-art PSNR and SSIM scores while maintaining higher computational efficiency compared to conventional Transformer-based approaches. Our work establishes a strong foundation for future research in adverse weather image restoration, offering a practical and effective solution for real-world applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.