基于大气散射物理模型的自适应雾图大气光强预测算法研究
Research on an Adaptive Prediction Algorithm for Atmospheric Light Intensity in Hazy Images

针对传统单幅图像去雾方法中大气光强度通常采用单一全局参数进行估计，难以适应复杂场景下非均匀光照与雾气分布，容易造成去雾图像整体偏暗、局部过曝光及色彩失真等问题，本文提出一种基于自适应大气光强预测的单幅图像去雾方法。该方法将大气光建模为空间变化的分布，通过构建空间自适应大气光估计网络，实现大气光参数的动态预测。在SOTS-outdoor和HSTS两个公开数据集上进行了定量与定性实验，实验结果表明，本文方法相较于传统 DCP 方法均表现出更好的图像恢复精度和结构保持能力。基于自适应大气光预测能够有效改善传统方法产生的暗色伪影和色彩失真问题，在复杂非均匀光照场景下具有更稳定的去雾效果。
To address the limitations of traditional single-image dehazing methods, which typically estimate atmospheric light intensity using a single global parameter and therefore have difficulty adapting to non-uniform illumination and haze distributions in complex scenes, often resulting in overall darkened dehazed images, local overexposure, and color distortion, this paper proposes a single-image dehazing method based on adaptive atmospheric light intensity prediction. The proposed method models atmospheric light as a spatially varying distribution and dynamically predicts atmospheric light parameters by constructing a spatially adaptive atmospheric light estimation network. Quantitative and qualitative experiments were conducted on two public datasets, SOTS-outdoor and HSTS. The experimental results demonstrate that the proposed method outperforms the traditional DCP method in terms of both image restoration accuracy and structural preservation. The adaptive atmospheric light prediction effectively alleviates dark artifacts and color distortion caused by traditional methods, providing more stable dehazing performance in complex scenes with non-uniform illumination.