基于双窗口对比和自适应容差的图像去雾  

Image dehazing based on dual window contrast and adaptive tolerance

作　　者：赵艳锋[1] 杨立东[1] 郭勇[1] 唐志斌 Zhao Yanfeng;Yang Lidong;Guo Yong;Tang Zhibin(School of Information Engineering,Inner Mongolia University of Science&.Technology,Baotou 014010,China;Beijing JENSEC Technology Limited Company,Beijing 102600,China)

机构地区：[1]内蒙古科技大学信息工程学院,包头014010 [2]北京简易网安科技有限公司,北京102600

出　　处：《国外电子测量技术》2022年第12期13-19,共7页Foreign Electronic Measurement Technology

基　　金：国家自然科学基金面上项目(81871430);内蒙古自治区自然科学基金(2021MS06030)项目资助。

摘　　要：雾天环境下图像质量下降,给户外视觉系统的应用带来极大阻碍。为解决此问题,提出一种基于双窗口对比和自适应容差的图像去雾算法。首先通过两个窗口逐像素对比来改进暗通道,使用改进后的暗通道求取透射率;然后提出自适应容差机制修正透射率,有效解决明亮区域透射率估计不准确的问题;最后将图像变换到HSV色彩空间,通过增强V分量来提高图像整体亮度。实验表明,算法能够较好地解决伪影和失真问题,在数据集上,峰值信噪比值为18.31、结构相似性值为0.81、信息熵值为7.52、平均梯度值为7.34。与传统和深度学习方法对比均有显著优势,相较于新颖的深度学习算法,4项指标分别提升了46.59%、14.08%、2.45%和35.42%。Image quality degrades in foggy weather, which greatly hinders the application of outdoor vision systems. To solve this problem, we propose an image dehazing method based on dual window contrast and adaptive tolerance. First, the dark channel is improved by pixel-by-pixel comparison of two windows, and the transmittance is calculated using the improved dark channel. Then, an adaptive tolerance mechanism is proposed to correct the transmittance, which effectively solves the problem of inaccurate transmittance estimation in bright areas. Finally, the image is transformed to HSV color space, and the overall brightness of the image is enhanced by the V component. Extensive experiments show that our algorithm can solve the problems of artifacts and distortion. On the dataset, the peak signal to noise ratio is 18.31, the structural similarity index metric is 0.81, the information entropy is 7.52, and the average gradient is 7.34. It has significant advantages in the comparison of traditional and deep learning methods. Compared with the new deep learning algorithm, the four indicators have increased by 46.59%, 14.08%, 2.45% and 35.42% respectively.

关 键 词：图像去雾 暗通道先验 窗口对比 自适应容差 色彩空间 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]