Abstract: Dark channel prior often keeps residual haze near depth edges after haze removal, and leads to a dim image and color distortion in sky area. Moreover, its processing speed is too slow. To solve these problems, this paper proposes an effective and fast method for real-time video dehazing based on dark channel prior and guided filter. Firstly, this paper replaces patch-based dark channel prior into pixel-based dark channel prior to eliminate block effect, applies guided filter to refine the transmission map, and employs quad-tree subdivision to estimate atmospheric light precisely. Furthermore, in order to get a visually-pleasing result, this paper adopts histogram equalization to enhance dehazed image. Finally, this paper uses downsampling and redundant information between each frame to reduce computational burden. Experimental results show that the proposed method is much better, it has less calculation quantity and a wide range of application suitable for real-time video dehazing.

Key words: video dehazing, pixel-based dark channel prior, quad-tree subdivision, guided filter, histogram equalization

摘要： 针对暗原色先验算法出现的边缘残雾、天空区域彩色失真、去雾后图像偏暗以及实时性差等问题，提出了一种基于点暗原色先验和引导滤波的视频去雾算法。采用逐点式最小值滤波来消除块效应，并利用四叉树法来快速准确地估计大气光值，结合直方图均衡化技术来增强图像，改善视觉效果，同时利用图像采样技术和引导滤波优化算法提高速度。实验结果显示，该算法的去雾图像清晰，运算量小，适用范围广，鲁棒性好，适合实时视频去雾。

关键词: 视频去雾, 点暗原色先验, 四叉树法, 引导滤波, 直方图均衡化