Scholarlink

<p>The underwater environment is characterized by uncertainties, including light attenuation, refraction, scattering, and various underwater media, which lead to color imbalance and loss of details in underwater images. Additionally, the number of paired underwater images available for network training is extremely limited. To address these challenges, we propose a novel end-to-end underwater image enhancement method called CDME. The method consists of three main components: the CDC-Net for color-depth correction, the MSR-Net for handling uneven lighting and color bias, and a custom Encoder–Decoder Network for final enhancement. In this CDC-Net, we introduce different depth information for each image as a clue. We propose a dynamic depth module to ensure that feature learning at different stages of our method fully incorporates the added depth information. The Module incorporating depth information are outputted as advanced cues to guide color correction. To further improve the quality of color-corrected underwater images, we perform both traditional static algorithm-based enhancement and dynamic deep learning-based enhancement. Finally, extensive experimental results on benchmark datasets demonstrate that CDME exhibits superior performance. The dataset and demo code are available at <a href="https://github.com/qwertyuiasdbvezdr/CDME">https://github.com/qwertyuiasdbvezdr/CDME</a>.</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

Depth-guided color correction and multi-scale Retinex network for underwater image enhancement

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Zhan Hu,
  • Juan Zhang,
  • Yongbin Gao,
  • Bo Huang,
  • Zhijun Fang

摘要

The underwater environment is characterized by uncertainties, including light attenuation, refraction, scattering, and various underwater media, which lead to color imbalance and loss of details in underwater images. Additionally, the number of paired underwater images available for network training is extremely limited. To address these challenges, we propose a novel end-to-end underwater image enhancement method called CDME. The method consists of three main components: the CDC-Net for color-depth correction, the MSR-Net for handling uneven lighting and color bias, and a custom Encoder–Decoder Network for final enhancement. In this CDC-Net, we introduce different depth information for each image as a clue. We propose a dynamic depth module to ensure that feature learning at different stages of our method fully incorporates the added depth information. The Module incorporating depth information are outputted as advanced cues to guide color correction. To further improve the quality of color-corrected underwater images, we perform both traditional static algorithm-based enhancement and dynamic deep learning-based enhancement. Finally, extensive experimental results on benchmark datasets demonstrate that CDME exhibits superior performance. The dataset and demo code are available at https://github.com/qwertyuiasdbvezdr/CDME.