An improved image dehazing method is proposed to address the issues of blurred details and color distortion in the restored image obtained by traditional dehazing algorithms. Firstly, the main structural image and the minimum channel image of the hazy image are acquired. The brightness mapping of the main structural image is calculated, and the maximum and sub-maximum points are identified to form four candidate regions. The median value of the region with the minimum entropy is selected as the global atmospheric light value. Subsequently, a fog line reliability evaluation parameter is introduced to determine whether transmission rate points belong to the noise area. Clustering of transmission rate points in the noise area is conducted, and the transmission rate is optimized using neighborhood fog lines. Clusters with too few pixels are merged, and the selection range of the maximum irradiance is appropriately expanded to compensate for errors caused by limited areas. Finally, edge information in the minimum channel is extracted using a side window box filter. An adaptive weight factor based on the characteristics of fog line clustering results is designed to remove texture information, further refine the transmission rate, and ultimately obtain the restored image based on the atmospheric imaging model. Experimental results demonstrate that the proposed algorithm shows significant performance improvements compared with various dehazing algorithms, with enhancements in information entropy, average gradient, blur coefficient, and fog concentration evaluation index (FADE), resulting in more complete details and better color matching with human visual perception in the restored image.

针对去雾算法得到的复原图像细节模糊和颜色易失真问题, 提出一种改进的图像去雾方法。首先, 获取含雾图像的主结构图及最小通道图, 计算主结构图像的行列亮度映射, 并寻找最值和次最值点, 构成4个备选区域, 将其中最小熵区域的中值作为全局大气光值。然后, 引入雾线可靠性评估参数来判断透射率点是否属于噪声区域, 对噪声区域的透射率点进行聚类, 并利用邻域雾线优化透射率来合并像素点过少的簇, 适当放大最大辐照度的选择范围来弥补区域受限带来的误差。最后, 使用侧窗盒式滤波, 提取最小通道中的边缘信息, 并根据雾线聚类结果的特性设计相对总变分的自适应权重因子, 去除纹理信息, 进一步细化透射率, 最终根据大气成像模型得到复原图像。实验结果表明, 与其他去雾算法相比, 所提算法在复原图像的信息熵、平均梯度、模糊系数及雾感知密度评估参数(FADE)等指标上均显著改进, 复原图像的细节更完整, 色彩能够更好地匹配人类视觉感知。