Scholarlink

<p>Underwater images often suffer from color deviation and low contrast due to scattering and absorption. Existing methods rely heavily on spatial-domain processing, inadequately utilizing frequency-domain information and lacking the integration of multi-scale frequency-domain decomposition with efficient sequence modeling tailored for underwater characteristics. To address these limitations, we propose a novel underwater image enhancement (UIE) network called Wavelet-Mamba underwater visual enhancement (WM-UVE) network, which pioneers the deep fusion of the wavelet domain and Mamba for UIE. Specifically, 1) wavelet decomposition is employed to decouple the image into low-frequency structure and high-frequency detail components; 2) the Mamba fusion scan low-frequency (MFSLF) processing module effectively restores underwater color distortion using a long-range state-space model; and 3) the shifted window-based high-frequency enhancement (SwinHF) module enhances high-frequency textures through a dynamic window attention mechanism. The proposed method has been validated on multiple datasets, demonstrating competitive performance across the board.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Underwater image enhancement via multi-scale frequency-domain decomposition and state-space modeling

  • 期刊论文

  • 发表日期：

    2026年4月3日

查看全文
  • Yinghao Zhang,
  • Jiamin Hu,
  • Haiyuan Cui,
  • Jieru Chi,
  • Guowei Yang,
  • Teng Yu

摘要

Underwater images often suffer from color deviation and low contrast due to scattering and absorption. Existing methods rely heavily on spatial-domain processing, inadequately utilizing frequency-domain information and lacking the integration of multi-scale frequency-domain decomposition with efficient sequence modeling tailored for underwater characteristics. To address these limitations, we propose a novel underwater image enhancement (UIE) network called Wavelet-Mamba underwater visual enhancement (WM-UVE) network, which pioneers the deep fusion of the wavelet domain and Mamba for UIE. Specifically, 1) wavelet decomposition is employed to decouple the image into low-frequency structure and high-frequency detail components; 2) the Mamba fusion scan low-frequency (MFSLF) processing module effectively restores underwater color distortion using a long-range state-space model; and 3) the shifted window-based high-frequency enhancement (SwinHF) module enhances high-frequency textures through a dynamic window attention mechanism. The proposed method has been validated on multiple datasets, demonstrating competitive performance across the board.