Scholarlink

<p>The marine ecosystem is crucial for sustaining life on Earth, and monitoring its flora and fauna is essential to detect changes and mitigate ecological threats. Regular assessments enable mapping of key elements and identification of anomalies. However, obtaining high-quality underwater images remains challenging due to marine snow and occlusion. This study addresses underwater image enhancement by evaluating Retinex-based algorithms, Multiscale Retinex with Colour Restoration (MSRCR) and Multiscale Retinex with Chromaticity Preservation (MSRCP), alongside two deep learning models, Water-Net and UWCNN++. Analysis is performed on underwater images of artificial reefs near the Port of Leixões, Portugal. A qualitative assessment through visual inspection and a quantitative evaluation using UCIQE, CCF, and UIQM are presented. Results highlight the strengths of Retinex-based approaches, with MSRCR achieving the best UCIQE (30.14) and CCF (0.53), and MSRCP leading in UIQM (6.56). Despite this, WaterNet was able to compete in color restoration, producing images that appeared visually realistic.</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

Comparative study of retinex algorithms and deep models for underwater image enhancement

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Tiago F. R. Ribeiro,
  • José Areia,
  • Bianca Reis,
  • João N. Franco,
  • Fernando Silva,
  • Rogério Luís de C. Costa

摘要

The marine ecosystem is crucial for sustaining life on Earth, and monitoring its flora and fauna is essential to detect changes and mitigate ecological threats. Regular assessments enable mapping of key elements and identification of anomalies. However, obtaining high-quality underwater images remains challenging due to marine snow and occlusion. This study addresses underwater image enhancement by evaluating Retinex-based algorithms, Multiscale Retinex with Colour Restoration (MSRCR) and Multiscale Retinex with Chromaticity Preservation (MSRCP), alongside two deep learning models, Water-Net and UWCNN++. Analysis is performed on underwater images of artificial reefs near the Port of Leixões, Portugal. A qualitative assessment through visual inspection and a quantitative evaluation using UCIQE, CCF, and UIQM are presented. Results highlight the strengths of Retinex-based approaches, with MSRCR achieving the best UCIQE (30.14) and CCF (0.53), and MSRCP leading in UIQM (6.56). Despite this, WaterNet was able to compete in color restoration, producing images that appeared visually realistic.