基于改进梯度算法的液晶屏亚像素级缺陷快速检测  

Fast detection of sub-pixel defects in liquid crystal display based on improved gradient algorithm

作　　者：钱基德[1,2] 陈斌[2,3] 钱基业[4] 王佐才[1,2] 

机构地区：[1]中国科学院成都计算机应用研究所,成都510041 [2]中国科学院大学,北京100049 [3]中国科学院广州电子技术有限公司,广州510070 [4]国网重庆市电力公司电力科学研究院,重庆401123

出　　处：《计算机应用》2017年第A01期201-205,共5页journal of Computer Applications

基　　金：四川省战略性新兴产业自主创新研发项目(SC2013510107033);四川省科技厅科技成果转化项目(2014CC0043);重庆市博士后科研项目(Xm2016060)

摘　　要：针对目前移动设备端的液晶屏亚像素级液晶颗粒缺陷检测效率低的问题,提出一种采用机器视觉的基于改进梯度的缺陷快速检测方法。首先,通过高分辨率的线阵相机获取超高分辨率的液晶屏的采集图,采集像素超过2亿,将微观的亚像素级液晶颗粒通过高分辨率相机转换成图像中的宏观目标;其次,提出一种自适应参数估计方法检测液晶屏中液晶颗粒排列参数,以满足不同规格的液晶屏的自适应缺陷检测;最后,基于获取到的液晶颗粒参数采用改进的梯度算法对采集图进行缺陷特征提取,采用优化中值滤波算法对获得的缺陷特征图像去噪,通过二值图像连通域标记优化算法对缺陷进行快速定位。理论与实验均表明该算法能够快速准确检测出液晶屏中的亚像素级液晶颗粒坏点,对生产检测中的偏转以及图像光照不均都有很好的鲁棒性,满足液晶屏生产流水线的检测需求。For low efficiency of sub-pixel liquid crystal particles defect detection on Liquid Crystal Display（ LCD）, a fast detection method based on improved gradient was proposed using machine vision. First, the high-resolution line array camera was used to capture the ultra-high resolution LCD images with more than 200 million pixels, and was used to convert the subpixel liquid crystal particles into the macro objects. Then, an adaptive parameter estimation method was proposed to detect the alignment parameters of the liquid crystal particles adapting to different LCD. Finally, the improved gradient algorithm was used to extract the defect features using the alignment parameters of the liquid crystal particles, the noises of the feature image were removed by the optimized median filter, and the defects were quickly located by the optimization algorithm for binary image connected component labeling. The theoretical analysis and experimental results show that the algorithm can detect the sub-pixel liquid crystal particle defects quickly and accurately, and has good robustness to the deflection of production detection and image illumination unevenness, and can meet the requirements of LCD production line detection.

关 键 词：液晶屏 缺陷检测 梯度 亚像素 机器视觉 高分辨率 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]