DATDM: Dynamic attention transformer diffusion model for underwater image enhancement

作者：Wang Hu, Shitu Chen, Tuyan Luo, Lijun Zhang, Hongjin Zhang, Zhixiang Liu, Shiwen Zhang, Jingxiang Xu · 发表于：Alexandria Engineering Journal · 年份：2025 · DOI：10.1016/j.aej.2025.04.078 · 被引用次数：7 · 研究领域：Image Enhancement Techniques、Image and Signal Denoising Methods、Advanced Image Processing Techniques

Underwater image degradation presents complex challenges, significantly impairing the efficiency of underwater tasks. The mainstream underwater image enhancement (UIE) methods are roughly divided into traditional physics-based and data-driven deep-learning methods. However, physical model-based methods have poor generalization capabilities. Data-driven methods require high-quality training data and suffer from limited model stability. We propose a novel UIE framework, DATDM, to address the above limitations and improve degraded underwater images. We first introduce a color correction module (CCM) to strengthen the model’s image restoration capabilities. On the other hand, we propose a novel dynamic attention transformer (DAT) denoising network with excellent performance. The proposed DAT denoising network utilizes a dynamic attention transformer mechanism that adaptively extracts feature information based on the complexity of the feature map, effectively capturing rich features while optimizing computational efficiency. The proposed DATDM method significantly outperforms the existing state-of-the-art methods, with peak signal-to-noise ratios of 24.05 and 27.39 and structural similarity index measures of 0.9233 and 0.9504 on the UIEB and LSUI datasets, respectively. The final experiments demonstrate that our DATDM achieves better performance and visual effects on UIE tasks.