针对当前去雾算法未考虑交通监控图像中雾气浓度分布不均匀的问题, 提出了波长相关物理成像模型的去雾算法.首先, 根据波长与雾气浓度的相关性, 构建了适用于交通监控图像的波长相关物理成像模型.然后, 根据波长与颜色的相关性, 设计出基于最大模糊相关图割的透射率估计算法.考虑到灰度值存在交叉重叠的模糊特性, 及景物的空间相关性, 利用递推的最大模糊相关算法快速获取景物划分信息, 并用此信息设计图割的数据项, 实施图割.该策略将基于阈值的分割算法与基于空间相关性的图割算法相结合, 确保了景物的空间相关性, 提高了分割精度, 避免白色目标的误分.最后, 通过分割结果中的天空区域, 准确地计算大气光, 实施去雾.在500幅仿真图像及真实图像上的测试结果表明, 该算法较已有去雾算法的去雾精度至少提高7%, 运行时间至少缩短了约15%, 可用于交通监控系统的图像去雾处理中.

Conventional dehazing algorithms usually neglect the problem of uneven fog concentration, which is existed in the traffic monitoring image. To tackle this problem, a dehazing algorithm based on wavelength related physical imaging model is proposed. Firstly, a wavelength related physical imaging model is built for traffic monitoring imaging process, in terms of the correlation between wavelength and fog concentration. Secondly, according to the correlation of color and wavelength, a transmission estimation strategy based on the maximal fuzzy correlation segmentation is designed. Considering the fuzzy property of overlapping intensities and spatial correlation of scenes, maximal fuzzy correlation is used for obtaining partition information of different scenes which helps to design the data term of graph cut and enable the implementation of graph cut. Such an image segmentation strategy incorporates the thresholdbased segmentation into the spatialbased segmention, ensuring the spatial correlation, improving the precision and further avoiding misclassfication of white object. Finally, the atmospheric light can be predicted by the sky region in the segmentaion result, and haze can be removed. The experiments tested on the 500 synthetic images and realworld images demonstrate that the proposed algorithm can improve dehazing precision by 7% at least and shorten running time by 15% roughly, comparing with exising dehazing algorithms. Hence, the proposed method can be used for image dehazing in the traffic monitoring system.