暗通道融合亮通道优化的夜间图像去雾算法  

Nighttime dehazing algorithm of dark channel and bright channel fusion optimization

作　　者：李鹏飞 何小海[1] 卿粼波[1] 石伟 吴小强[1] LI Peng-fei;HE Xiao-hai;QING Lin-bo;SHI Wei;WU Xiao-qiang(College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China;The Second Research Institute of Civil Aviation Administration of China, Chengdu 610000, China)

机构地区：[1]四川大学电子信息学院,四川成都610065 [2]中国民航局第二研究所,四川成都610000

出　　处：《液晶与显示》2021年第4期596-604,共9页Chinese Journal of Liquid Crystals and Displays

基　　金：成都市产业集群协同创新项目(No.2016-XT00-00015-GX);四川省科技计划(No.2018HH0143)。

摘　　要：夜间有雾环境中拍摄的图片具有光照不均、对比度低且模糊的现象。本文通过分析夜间有雾图像的成像特点,提出一种有效的夜间图像去雾方法。首先,在估计大气光时使用加权差分图像作为参考进行引导滤波来估计含轮廓信息的大气光;使用拉普拉斯锐化灰度图像作为参考引导滤波来估计补充细节的大气光,然后将算出的两种大气光融合得到最终的大气光。最后,基于广泛使用的雾霾场景成像模型,本文算法结合暗通道和亮通道构建了双通道联合优化的透射率函数,在计算透射率的过程中加入了自适应权重系数、限制对比度自适应直方图均衡化和引导滤波来提高透射率函数的精度。实验结果表明,在测试图片上计算的信息熵平均值为7.5872,对比度均值为23.8097,和对比算法相比均有一定提升。所提算法去雾结果视觉效果良好,在去雾的同时可以有效减少细节的损失。Images taken in night hazy environment usually have uneven illumination,low contrast and blurry phenomena.By analyzing the imaging characteristics of night hazy images,An effective nighttime dehazing algorithm is proposed in this paper.First,in estimating atmospheric light,weighted channel difference image is used as a reference to conduct guided filtering to estimate atmospheric light containing contour information.A Laplace-sharpened grayscale image is used as a reference to conduct guided filtering to estimate the atmospheric light supplementing the details,and then the two calculated atmospheric lights are combined to obtain the final atmospheric light.Finally,based on the widely used haze scene imaging model,the dark channel and light channel are combined to construct the transmission function of dual channel joint optimization,and in the process of estimating transmittance,adaptive weight coefficient,contrast limited adaptive histogram equalization and guided filtering are added to improve the accuracy of transmittance function.The experimental results show that the average information entropy calculated on the test pictures is 7.5872,and the average contrast is 23.8097,which are better than the comparison algorithms.The dehazing results of the proposed algorithm have a good visual effect and can effectively reduce the loss of details.

关 键 词：加权差分 拉普拉斯锐化 暗通道 亮通道 自适应权重系数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]