结合暗通道的自适应MSRCR图像去雾算法

Adaptive MSRCR image defogging algorithm combined with dark channel

针对传统Retinex图像增强算法在图像去雾中无法对雾天图像实际情况进行去雾,导致增强后的图像出现光晕效应和细节丢失等问题,提出一种结合暗通道先验理论和引导滤波方法的自适应的多尺度Retinex图像去雾算法。利用子块平移部分重叠法将雾天图像进行分块处理;通过分析雾天图像的透射率和引导滤波尺度的相关性,利用最小二乘法构建自适应尺度函数;采用带自适应效果的引导滤波估计具有结构信息的光照分量,以此来还原雾天图像的反射分量;最后分析了增强后的雾天图像出现颜色失真的原因,并进行色彩校正。实验结果表明,相较于其他算法,本算法分别在信息熵、可视边缘梯度和新增可见比均有较大提升,对不同景深的雾天图像均具有良好的增强效果。

Aiming at the problem that traditional Retinex image enhancement algorithm can not defog actual situation of foggy image, which leads to halo effect and loss of details in enhanced image, an adaptive multi-scale Retinex image defogging algorithm combining dark channel prior theory and guided filtering method is proposed. The sub-block translation partial overlap method is used to segment fog image;By analyzing the correlation between the transmittance and the scale of guided filter, an adaptive scale function is constructed by using the least square method;The improved guided filter with adaptive effect is used to estimate illumination component with structure information, to restore reflection component of fog image;Finally, the causes of color distortion in enhanced fog image are analyzed and color correction is carried out. The experimental results show that, compared with other algorithms, the proposed algorithm improves information entropy, visual edge gradient, and new visibility ratio, and has a good enhancement effect for fog images with different depths of field.