基于暗原色先验的快速视频去雾优化算法

Fasting video haze removal algorithm using dark channel prior

对目前高分辨率视频图像去雾算法实时性差,天空及大量明亮区域处理不理想等问题,提出一种基于暗原色先验的快速视频去雾优化算法。针对视频图像,采用导向滤波和帧差法,实现快速视频去雾。根据经典大气散射物理模型,首先,利用暗原色先验估计大气光值和透射率图;然后,下采样透射率图并用导向滤波得到优化的透射率图后,上采样并改善透射率图;最终,得到去雾视频帧。与带颜色恢复的多尺度Retinex算法(MSRCR)和He算法进行对比,实验结果表明,所提优化算法能有效提高视频去雾速度,改善去雾效果。

To overcome the defects of the existing algorithms in the high resolution video, such as the poor re- al-time performance, bad effect in sky area dehazed image, a fasting video haze removal algorithm which is based on the dark channel prior was proposed. For video images, the frame difference method and guided fil- tering are combined to fast video defogging. According to the classical atmospheric scattering physical model, the dark channel prior is used to estimate atmospheric light and initial transmittance. Secondly, the method of guided filtering was used to refine the down-sampled rough transmission map. refined transmission map was upsampled and corrected to obtain the final transmission map. Finally, the clear video frames were got. Com- parison experiments with other two kind of methods were given, including MSRCR （ Multi-Scale Retinex with Color Restoration） and He. The experimental results show that the proposed method can improve the process- ing speed of video defogging, and the quality of video defogging is improved.