Scholarlink

<p>Underwater image enhancement is a crucial technique for improving the quality of underwater photographs, widely applied in fields such as underwater target tracking, robotics, and biological exploration. The unique characteristics of underwater environments often lead to issues like reduced contrast, color distortion, loss of detail, and blurred textures. Therefore, our proposed OVMamba method leverages a global omnidirectional state space to enhance the capture of global information, improving image quality by modeling relationships between pixels and channels. To tackle inconsistent quality, we introduce a multi-head channel attention mechanism, focusing on different subspaces and channels, particularly those with significant quality degradation, to enhance contrast. Additionally, we perform feature extraction independently within RGB color channels and adaptively adjust channel weights to restore natural colors and enhance visual perception. Experimental results demonstrate the effectiveness of our method across multiple datasets.</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

OVMamba: omnidirectional vision state space model for underwater images enhancement

  • 期刊论文

  • 发表日期：

    2025年5月9日

查看全文
  • Yufeng Li,
  • Kang Liu,
  • Zitian Zhao

摘要

Underwater image enhancement is a crucial technique for improving the quality of underwater photographs, widely applied in fields such as underwater target tracking, robotics, and biological exploration. The unique characteristics of underwater environments often lead to issues like reduced contrast, color distortion, loss of detail, and blurred textures. Therefore, our proposed OVMamba method leverages a global omnidirectional state space to enhance the capture of global information, improving image quality by modeling relationships between pixels and channels. To tackle inconsistent quality, we introduce a multi-head channel attention mechanism, focusing on different subspaces and channels, particularly those with significant quality degradation, to enhance contrast. Additionally, we perform feature extraction independently within RGB color channels and adaptively adjust channel weights to restore natural colors and enhance visual perception. Experimental results demonstrate the effectiveness of our method across multiple datasets.