基于暗原色先验的雾霾天气图像清晰化算法

A clearness algorithm for smog images based on dark channel prior

传统基于暗原色先验的去雾算法稳定且去雾效果好,但算法运行时间长。为在运算时间和图像清晰度之间取得平衡,在传统基于暗原色先验的去雾算法基础上,提出一种新的雾霾天气图像清晰化算法。算法在求原始有雾图像的暗原色时,将传统基于暗原色先验算法中固定的局部区域大小设置为随图像大小变化的值,从而增强算法的自适应性;设置大气光值的阈值,避免大气光估计值过高造成的去雾后图像整体向白场过渡;采用导向滤波算法代替传统基于暗原色先验算法中的软抠图算法,提升算法的运行效率;最后利用自动色阶算法调整去雾后图像颜色的明暗分布。实验结果表明,该算法输出的图像对比度、清晰度好,色彩保真度高,明暗分布合理,算法稳定且运行时间短,实现了清晰度和运算时间的平衡。

The traditional de-hazing algorithm based on dark channel prior is stable and has good haze removal effect,but its running time is very long.To balance the operation time and the clearness of the image,we propose a new clearness algorithm for smog images based on the traditional dark channel prior based de-hazing algorithm.The clearness algorithm replaces the fixed size of local area in the traditional dehazing algorithm by a changing value that varies with the size of the image when seeking the dark channel of the original smog image,thus enhancing the adaptability of the algorithm.We set a threshold for the atmospheric light to prevent the overall image from transitioning to the white field due to the too high estimated value of atmospheric light.We use the guided filtering algorithm instead of the soft matting algorithm to improve algorithm efficiency.Finally,we employ the auto-color algorithm to adjust the color lighting distribution of the dehazed image.Experimental results show that the image outputted by the proposed algorithm has good contrast and clearness,high color fidelity,and reasonable lighting distribution.Besides,the algorithm is stable and has short running time,thus realizing a balance between running time and clearness of the image.