大气散射模型的物理机制及其对单图像去雾的理论指导研究

The Physical Mechanism of Atmospheric Scattering Model and Its Theoretical Guidance for Single Image Dehazing

摘要: 单幅图像去雾在计算机视觉的高层任务中具有重要意义。本文以大气散射模型为核心，系统梳理了去雾技术的理论基础与发展脉络。首先阐述了光传播的物理机制与去雾的病态逆问题本质；其次，对比分析了基于先验知识的传统参数估计方法与基于深度学习的端到端范式，并重点探讨了无监督与零参考框架在解决真实场景域偏移中的优势；最后，针对非均匀雾霾与零参考先验缺失等难题，提出了非均匀物理模型拓展、物理模型引导的微观网络轻量化以及物理规律正则化等优化方向。本文旨在为去雾技术的物理可解释性与工程落地提供理论参考。

Abstract: Single image dehazing is of profound significance for high-level computer vision tasks. Centered on the atmospheric scattering model, this paper systematically reviews the theoretical foundation and technological evolution of dehazing methods. First, the physical mechanism of light propagation and the ill-posed nature of the dehazing problem are elaborated. Second, traditional prior-based parameter estimation methods and deep learning-based end-to-end paradigms are compared, highlighting the advantages of unsupervised and zero-reference frameworks in addressing domain shift in real-world scenarios. Finally, focusing on challenges such as non-uniform haze and the lack of zero-reference priors, future optimization directions including the expansion of non-uniform physical models, micro-network lightweight design guided by physical models, and physical law regularization are proposed. This paper aims to provide a theoretical reference for the physical interpretability and practical deployment of dehazing technologies.