一种自适应暗通道先验去雾算法

基于暗通道先验的图像去雾算法是一种简单有效的图像去雾算法，但该算法在处理较高分辨率的图像时，时间复杂度较高，复原后的图像亮度偏低，且在处理具有大面积白色明亮区域图像时存在色彩失真的问题。针对这些问题，本文提出了改进的自适应暗通道先验去雾算法，新算法引入自适应的指导滤波法代替原算法中的软抠图法，提高算法的计算效率的同时获得最优滤波窗口半径。同时，新算法还通过改进透射率图估计方法，弱化对明亮区域的去雾处理，避免过增强，并调整图像亮度，优化去雾结果。通过合成雾图和真实场景雾图实验验证了新算法的有效性。合成雾图实验中采用全参考评价方式，在清晰的无雾场景上模拟雾的形成，计算加雾前与使用不同算法去雾后图像的绝对差值进行比较。真实场景雾图实验中，采用基于人类视觉感知的CNC(Contrast-naturalness-colorfulness)综合评价体系，计算同一雾图在不同算法去雾后图像的CNC指数。实验结果表明，在相同图像分辨率条件下，本文提出的自适应算法不仅去雾后图像视觉效果更加理想，而且处理时间大为减少。

An Adaptive Haze Removal Algorithm Based on Dark Channel Prior

The image haze removal algorithm based on dark channel prior is simple and effective. However, this algorithm has high time complexity for high resolution images, low brightness for the results of restoration, color distortions for the images with large high-light area. To solve these problems, a modified adaptive algorithm based on dark channel prior was proposed for single image haze removal in this paper. In the proposed novel algorithm, the adaptive guided image filtering algorithm was adopted to instead of soft matting for increasing the computing efficiency and obtaining the optimized radius of filtering window at the same time. Moreover, adaptive transmission and brightness adaption were also applied to overcome color distortion problems and enhance the performances of the original algorithm. Both synthetic hazy images and real scene images were used for the validation of the proposed algorithm. The absolute different values (a full-reference way) and CNC (Contrast-naturalness- colorfulness) system (a no-reference way) were respectively used to assess the haze removal results for the synthetic hazy images and real scene images. The statistical results indicated that the proposed algorithm was more effective than the original algorithm according to the quantitative criteria as well as the human visual perception when the images have the same resolution and size.