水下非均匀光照场景下的混凝土图像增强方法  

Concrete image enhancement method for underwater uneven illumination scenes

作　　者：林海涛 王皓冉 李永龙 陈永灿 张华[1,3] LIN Haitao;WANG Haoran;LI Yonglong;CHEN Yongcan;ZHANG Hua(College of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,China;Sichuan Energy Internet Research Institute,Tsinghua University,Chengdu 610213,China;Sichuan Tianfu New Area Innovation Research Institute,Chengdu 610000,China;School of Civil Engineering and Geomatics,Southwest Petroleum University,Chengdu 610500,China)

机构地区：[1]西南科技大学信息工程学院,绵阳621010 [2]清华四川能源互联网研究院,成都610213 [3]四川天府新区创新研究院,成都610000 [4]西南石油大学土木工程与测绘学院,成都610500

出　　处：《清华大学学报（自然科学版）》2023年第7期1144-1152,共9页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(52009064,U21A20157);四川省科技厅项目(2022JDRC0073,2022YFSY0011,2022YFQ0080)。

摘　　要：检测水下基础设施混凝土表观缺陷是保障水下基础设施安全稳定运行的重要措施。使用遥控水下机器人(remote operated vehicle,ROV)采集混凝土表观图像是当前水下检测最高效的方式,然而ROV采集到的混凝土图像存在光照不均、色彩失衡、对比度差和边缘信息弱等问题。该文针对水下混凝土图像质量不佳的问题,提出了水下非均匀光照场景下的混凝土图像增强方法。首先采用图像修复技术(image inpainting technique,IIT)对图像局部高光区域进行修复;然后在暗通道图像增强方法的基础上引入图像对比度感知调节方法,选取不同窗口尺寸,在每个局部窗口区域中实现图像增强;最后采用自然图像质量评估、基于感知的图像质量评估、无参照物图像质量评估和水下彩色图像质量评估4个指标对增强后的图像进行评估。实验结果表明:该文提出的方法在多个对比指标中优于现有水下图像增强方法,能有效提升水下图像质量。[Objective]Identifying concrete surface defects in underwater infrastructure is crucial to ensure safe and stable operation.At present,using ROV to gather concrete surface images is the most effective measure for underwater image detection.However,the concrete images obtained by ROV have some phenomena,such as uneven illumination,color imbalance,poor contrast,and weak edge information.In this study,the issue of underwater concrete photos with poor image quality in nonuniform lighting scenarios is investigated,and a method for underwater concrete image enhancement is suggested,which provides efficient data support for the detection and analysis of concrete surface defects in underwater infrastructure.[Methods]The local highlight issue caused by the fill light spots in underwater images is processed based on image repair.First,the image is thresholded,and the highlighted pixel area of the image is retained.Second,using the hough circle detection method,the annular fill light spot area on the image is retrieved.The resulting annular fill light spot image is then used as the Mask image of the original image for restoration.Finally,the local highlighted area caused by the fill light spot is repaired by filling the adjacent pixel area.An improved dark channel prior(DCP)technique is used to enhance the image to address the issue of poor image quality brought by the undersea environment’s uneven illumination.However,the single window size will have the following three problems:(1)A small size window area will lead to the oversaturation of local areas in the enhanced image.(2)A large window size can better eliminate the haze of the image but may produce halos.(3)A single-sized window is difficult to adapt to different pixel size images.Therefore,the selection area of the dark channel window size is expanded upon in this work using the contrast perception method.Image enhancement is done in each local window region by computing the contrast of seven neighborhood windows of one pixel and choosing the relevant dark channel

关 键 词：混凝土图像增强 水下测量 图像畸变矫正 对比度引导 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]