###
工程科学与技术:2023,55(3):225-234
←前一篇   |   后一篇→
本文二维码信息
码上扫一扫!
一种基于自适应大气光和加权引导滤波的夜间图像去雾算法
(1.四川大学 机械工程学院,四川 成都 610065;2.四川省智能农机装备创新中心,四川 德阳 618000)
A Night Image Dehazing Algorithm Based on the Adaptive Atmospheric Light and the Weighted Guided Filter
(1.School of Mechanical Eng., Sichuan Univ., Chengdu 610065, China;2.Sichuan Center of Collaborative Innovation for Intelligent Agricultural Machinery, Deyang 618000, China)
摘要
图/表
参考文献
相似文献
附件
本文已被:浏览 485次   下载 473
投稿时间:2021-12-13    修订日期:2022-05-27
中文摘要: 夜间图像去雾对于夜间场景下无人驾驶、交通安防等有重要的工程应用价值。针对暗通道先验算法在夜间雾天场景下失效的问题,提出一种基于自适应大气光和加权引导滤波的夜间图像去雾算法。该算法首先基于图像亮度和饱和度联合求取信道图,并将信道图作为引导图对原图像进行引导滤波得到大气光分布图;为解决暗通道先验在图像亮区域的失效问题,引入亮通道先验矫正亮区域的透射率;为优化亮、暗通道透射率的融合,建立一种基于分段伽马矫正的融合权值计算方法,用于亮区域透射率的权值计算,并利用该透射权值加权得到图像的初始透射率。然后,利用加权聚合引导滤波代替引导滤波细化初始透射率,解决引导滤波弱化细小纹理而引起的边缘模糊问题。最后,将复原图像转换到HSV(Hue–Saturation–Value)色彩空间,对亮度分量V(Value)进行均衡化调整,并对均衡化前后的图像进行线性加权获得最终复原结果。实验结果表明:所提算法大气光分布图估值合理,可有效反映夜间多光源场景下的大气光分布情况;图像亮、暗区域透射率计算准确,复原图像去雾彻底、纹理清晰;与经典算法对比显示,复原结果的峰值信噪比、信息熵、平均梯度和方差的最大提升幅度分别为49.4%、18.3%、172.3%、115.0%,综合指标优于所对比的其他算法。
中文关键词: 夜间去雾  大气光  加权聚合  亮通道
Abstract:Nighttime image dehazing provides significant industrial contributions to unmanned driving and traffic security. To address the inability of the dark channel prior algorithm facing the foggy night scenes, a night image dehazing algorithm is proposed based on adaptive atmospheric light and weighted guided filtering. The distribution map of the atmospheric light is estimated using the guided filter based on the channel map obtained by image brightness and saturation. Bright channel prior is introduced to correct the transmittance of the bright region to solve the problem of dark channel failure in the bright region. To enhance the fusion of the light and dark channels, a weighted fusion method is used to calculate the weight of the bright region based on piecewise gamma correction, which is further applied to infer the initial transmittance of the raw image. A weighted aggregation guided filtering is proposed to refine the initial transmittance, focusing on addressing the edge blurring caused by the weakened tiny texture of the guided filtering. After converting the intermediate restored image into the HSV domain, image equalization is performed on the V (Value) component to further enhance the image. Finally, the linear weighting algorithm is applied to fuse the enhanced image and the intermediate restored image, to obtain the final recovered image. Experimental results demonstrate that the proposed algorithm is able to properly estimate the atmospheric light distribution map, which can effectively reflect the atmospheric light distribution for the multi-light source in a nighttime scene. The transmittance of both the bright and dark image regions can be accurately calculated to completely recover the raw image with clear texture. In addition, the proposed approach outperforms other comparative baselines, harvesting 49.4%, 18.3%, 172.3%, and 115.0% relative performance improvement on the peak signal-to-noise ratio, information entropy, average gradient, and variance of the restoration results, respectively.
文章编号:202101218     中图分类号:TP391.4    文献标志码:
基金项目:四川省重大科技专项(2020YFSY0058)
作者简介:第一作者:赵波赵 波(1972—),男,教授,博士. 研究方向:矿山特殊工况无人驾驶感知和定位. E-mail:zhaobo@scu.edu.cn
引用文本:
赵波,李洪平,金汝宁,唐万松,刘相宜.一种基于自适应大气光和加权引导滤波的夜间图像去雾算法[J].工程科学与技术,2023,55(3):225-234.
ZHAO Bo,LI Hongping,JIN Runing,TANG Wansong,LIU Xiangyi.A Night Image Dehazing Algorithm Based on the Adaptive Atmospheric Light and the Weighted Guided Filter[J].Advanced Engineering Sciences,2023,55(3):225-234.