水下图像增强技术:挑战、进展与未来方向

Underwater Image Enhancement Technology:Challenges,Progress and Future Directions

摘要:

水下图像是一种重要视觉信息，服务于海洋科学研究及资源勘探等领域。为从退化的低质水下图像中获得清晰化的高质图像，需要采取有效的图像增强技术。从水下环境的特殊性阐述了水下图像的退化机理及成像模型。综述了基于传统图像处理的水下图像增强和基于深度学习驱动的水下图像增强这两类主流技术，分析了基于物理模型的方法、基于非物理模型的方法、基于卷积神经网络的方法、以及基于生成式网络的方法存在的不足。归纳了现在水下图像增强技术存在的核心挑战和解决方案。并对未来的研究方向进行了展望，包括：物理模型与深度学习的深度融合；无监督/监督/自监督学习；水下图像生成与高质量数据集构建；多模态信息融合；轻量化、高效率与嵌入式部署；客观评价指标与主观感知一致性。

Abstract:

Underwater images represent a crucial form of visual information，serving fields such as marine scientific research and resource exploration. To obtain clear and high-quality images from degraded low-quality underwater images，effective image enhancement techniques are essential. In this paper，the degradation mechanisms and imaging models of underwater images are expounded based on the unique characteristics of underwater environment. Two mainstream techniques ， namely traditional image processing-based underwater image enhancement and deep learning-driven underwater image enhancement，are reviewed. The limitations of physical model-based methods，non-physical model-based methods，convolutional neural network-based methods，and generative network-based approaches are analyzed. Then，the core challenges and corresponding solutions in current underwater image enhancement technologies are summarized. Furthermore ， future research directions are prospected，including：deeper integration of physical models with deep learning; unsupervised/weakly-supervised/ self-supervised learning; underwater image generation and high-quality dataset construction; multi-modal information fusion; lightweight design，high efficiency and embedded deployment; and objective evaluation metrics aligned with subjective perceptual consistency.