融合平均曲率与直方图分析的暗通道先验去雾算法

A Dark Channel Prior Dehazing Algorithm Combining Mean Curvature and Histogram Analysis

针对当前暗通道先验算法存在的边缘模糊、过饱和以及大气光值错估问题,提出一种融合平均曲率与直方图分析的暗通道先验去雾算法。首先,计算输入图像微观几何表面上所有像素点的平均曲率,并以归一化的平均曲率强度为权重融合小尺度与大尺度传输图;其次,分析存雾图像三个通道的直方图,确定是否存在过亮区域并估计全局大气光值;最后,利用大气散射模型得到清晰的去雾图像。与其他算法在HAZERD等公共数据集上的实验对比结果表明,该算法能够解决当前暗通道算法存在的问题,且具有更高的鲁棒性与实效性。

A dark channel prior dehazing algorithm that combines mean curvature and histogram analysis is proposed to address the issues of edge blur,oversaturation,and misestimation of atmospheric light values in current dark channel prior algorithms.Firstly,calculate the mean curvature of all pixels on the micro geometric surface of the input image,and fuse the small-scale and large-scale transmission maps with the normalized mean curvature intensity as the weight;secondly,analyze the histograms of the three channels of the foggy image to determine whether there are overly bright areas and estimate the global atmospheric light value;finally,a clear dehazing image is obtained using the atmospheric scattering model.The experimental comparison with other algorithms on public datasets such as HAZERD shows that this algorithm can solve the problems of current dark channel algorithms and has higher robustness and effectiveness.