%0 Journal Article
%T 水下图像恢复算法综述<br>A Review of Underwater Image Restoration Algorithms
%A 李哲怡
%J Journal of Image and Signal Processing
%P 257-270
%@ 2325-6745
%D 2025
%I Hans Publishing
%R 10.12677/jisp.2025.142024
%X 水下图像恢复算法的研究旨在解决水下环境中由于光线吸收和散射导致的图像质量下降问题&#65292;以提高水下图像的清晰度和信息可用性。本综述总结了近年来水下图像恢复领域的研究进展&#65292;包括基于物理模型的方法、基于数据统计的方法以及基于深度学习的算法。物理模型方法通过分析水下光传输特性(如暗通道先验、红通道方法等)恢复图像清晰度&#65307;数据统计方法依赖于统计特性和优化技术(如直方图均衡、融合策略等)&#65307;基于深度学习的方法利用卷积神经网络(CNN)、生成对抗网络(GAN)等模型在无监督和有监督框架下实现高效的图像增强与恢复。这些方法在去雾、色彩校正和对比度增强方面取得了显著进展&#65292;但在应对复杂光学条件及处理速度优化方面仍存在挑战。综述最后探讨了该领域的发展趋势&#65292;指出结合物理模型与深度学习的混合方法以及实时恢复算法将成为未来的重要研究方向。<br>The study of underwater image restoration algorithms aims to address the quality degradation of underwater images caused by light absorption and scattering, thereby enhancing image clarity and usability for various applications. This review summarizes recent advancements in the field, focusing on three main categories: physics-based methods, data-based techniques, and deep learning-based approaches. Physics-based methods analyze light propagation characteristics in underwater environ- ments, employing techniques such as the dark channel prior, red channel methods, and polarization analysis to restore image clarity. Data-driven methods rely on statistical and optimization techniques, including histogram equalization, fusion strategies, and color correction models, to enhance image quality. Meanwhile, deep learning-based approaches utilize models like convolutional neural networks (CNNs) and generative adversarial networks (GANs) to achieve effective and adaptive restoration within both supervised and unsupervised frameworks. These methods have demonstrated remarkable advancements in dehazing, color correction, and contrast enhancement. Despite these achievements, challenges remain in addressing complex optical environments and improving computational efficiency for real-time applications. This review highlights the need for hybrid methods that integrate the strengths of physics-based models and deep learning, as well as the development of algorithms tailored for real-time underwater image restoration. Future research is expected to focus on achieving greater robustness, adaptability, and processing speed, enabling wider adoption in underwater robotics, marine research, and environmental monitoring.
%K 水下图像恢复&#65292
%K 图像增强&#65292
%K 物理模型&#65292
%K 深度学习&#65292
%K 实时处理<br>Underwater Image Restoration
%K Image Enhancement
%K Physics-Based Methods
%K Deep Learning
%K Real-Time Processing
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=111572