基于交互式的道路图像阴影去除  

Shadow Removal of Road Image based on Interactive

作　　者：邹志亮 周建华 张志波 刘俊文 郑子豪 Zou Zhiliang;Zhou Jianhua;Zhang Zhibo;Liu Junwen;Zheng Zihao(CATARC Automotive Test Center(Guangzhou)Co.,Ltd.,Guangzhou 511300)

机构地区：[1]中汽研汽车检验中心(广州)有限公司,广州511300

出　　处：《中国汽车》2023年第7期58-64,共7页China Auto

基　　金：广东省重点领域研发计划项目(2020B090905003)。

摘　　要：本文提出了一种交互式、鲁棒性强、高质量的快速阴影去除方法。首先,本文使用了即时学习的方法去除单幅图像阴影,该方法由两个阴影与光照区域像素的用户输入作为指导,由此得出用于阴影去除的初始阴影二值图像轮廓。其次,检测后的阴影图像再通过一个涂抹的颜色采样线指示阴影区域、阴影边界、亮度区域像素之间的照度梯度变化关系的即时学习方法来对图像进行偏移校正以消除环境光的影响。接着,对校正的图像进行聚类和照度估计。然后,计算无阴影区域统计分布和照明比例因子去除阴影。最后,使用光一致优化对去除阴影后的图像区域进行平滑处理和图像增强。本文使用公开的评估数据集定量和定性地证明了所提出的图像阴影去除方法简单有效,可以生成高质量的无阴影图像,并且能够处理具有丰富纹理类型和不均匀阴影的困难场景。提高了智能汽车视觉传感器的目标识别率。In this paper,an interactive,robust and high quality fast shadow removal method is proposed.Firstly,this paper uses a real-time learning method to remove the shadow of a single image.This method is guided by the user input of two shadows and pixels in the light⁃ing area,and thus obtains the initial shadow binary image contour for shadow removal.Secondly,the detected shadow image is corrected by the real-time learning method of the illuminance gradient relationship between pixels in the shadow region,shadow boundary and bright⁃ness region with a erased color sampling line to eliminate the influence of ambient light.Then,clusteringand illuminance estimation are car⁃ried out for the corrected images.Then,the statistical distribution of the shadow-free region is calculated and the illumination scaling factor is used to remove the shadow.Finally,the image region after shadow removal is smoothed and enhanced by optical uniformity optimization.Using publicly available evaluation data sets,we demonstrate quantitatively and qualitatively that the proposed shadow removal method is simple and effective,can generate high quality shadow free images,and can handle difficult scenes with rich texture types and uneven shad⁃ows.The object's recognition rate of intelligent vehicle vision sensor is improved.

关 键 词：彩色图像 阴影去除 阴影检测 即时学习 偏移校正 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]