摘要
针对单幅雾霾图像中包含的大面积天空或白色物体等区域暗通道先验失效和导向滤波去雾方法去雾不彻底的问题,提出了一种基于改进暗通道和导向滤波的单幅图像去雾算法.首先基于暗通道引入了混合暗通道,然后对混合暗通道进行映射处理,从而得到大气耗散函数粗估计值;利用导向滤波方法优化大气耗散函数粗估计值,进而求解环境光值和初始传输图;利用全变差正则化方法对初始传输图进行优化,以解决其平滑性较差的问题.实验结果表明,本文算法得到的去雾图像具有较高的清晰度,对于大面积天空或白色物体区域也能实现良好的去雾效果.
To tackle the problem that dark channel prior is invalid for large sky or bright objects regions of single hazy image and the problem that some hazes cannot be removed using guided filter, we propose a single image dehazing algorithm based on improved dark channel prior and guided filter. Firstly, we introduce a mixed dark channel based on dark channel. The coarser estimate of atmosphere veil is obtained after a map processing on the mixed dark channel.Then guided filter is utilized to optimize the coarser atmosphere veil. An initial transmission map can be obtained after getting more accurate atmospheric light with the optimized atmosphere veil. Finally, the total variation regularization method is utilized to address the problem of poor smoothness of the initial transmission map. Experimental results show that the recovered haze-free image with the proposed method has better sharpness, even for large sky or bright objects regions.
出处
《自动化学报》
EI
CSCD
北大核心
2016年第3期455-465,共11页
Acta Automatica Sinica
基金
国家自然科学基金(61471313)
河北省自然科学基金(F2014203076)资助~~
关键词
图像去雾
暗通道先验
导向滤波
全变差
Image dehazing
dark channel prior
guided filter
total variation
