Abstract: Concerning the shortcoming of low efficiency of the algorithm for single-image fog removal, an algorithm based on improved Kuwahara filter is proposed to detail rough transmission of foggy images. This algorithm introduces the square integral operator to transform calculation for pixel areas into calculation of four vertices of a rectangle. Amended algorithm based on threshold for transmission is used to go a step further detail the transmission for sky areas in order to increase the efficiency of the algorithm and avoid the color-cross of sky areas. Simulation test shows that it only needs 10 ms to deal with a foggy image of 800×600 pixel, and the velocity based on the improved Kuwahara is 3 000 times of He’s, 2 000 times of Shi’s, and could meet the real-time processing need for expressway images.

Key words: Fog; image processing, dark channel prior, Kuwahara filter, atmospheric model

摘要： 针对单幅图像去雾算法中存在运行效率低下的不足，提出改进的Kuwahara滤波细化粗糙透射率算法，引入积分图和平方积分图算法，将面积像素计算转化为目标矩形四个顶点的计算；对天空区域使用阈值透射率修正算法做进一步平滑，以提高算法执行效率，避免天空区域色彩失真。经仿真测试，对于800×600的雾图像，需10 ms即可完成去雾处理；对于大小为135 K的图像，改进Kuwahara处理速度是He的3 000倍，是Shi的2 000倍，表明算法能满足高速图像处理的实时性需求。

关键词: 雾, 图像处理, 暗原色, Kuwahara滤波, 大气散射模型