基于图像配准的STN-LCD外观缺陷检测-沈会良.pdfVIP

第35 卷第9 期 光电工程 Vol.35, No.9 2008 年 9 月 Opto-Electronic Engineering Sep, 2008 文章编号：1003-501X(2008)09-0060-06 基于图像配准的 STN-LCD 外观缺陷检测 沈会良，张宏刚，李志能 ( 浙江大学 信息与电子工程学系，杭州 310027 ) 摘要：提出了一种基于图像配准的超扭曲向列液晶显示器件(STN-LCD)的外观缺陷自动检测方法。该方法首先对 标准模板图像做不均匀光照消除、二值化以及区域信息提取；然后通过控制点检测和仿射变换实现待检测图像和 模板图像之间的配准；并利用各图形区域的灰度平均值和标准方差等统计信息，检测缺段、针孔等各类缺陷。为 提高图像配准精度，进一步提出了有效控制点筛选方案以及混合插值方法。实验结果表明，该方法设计思路合理， 缺陷检测正确率达到 98.3% ，可代替人眼实现对 STN-LCD 多种外观缺陷的快速、自动检测，满足实际应用需求。 关键词：STN-LCD；缺陷检测；图像配准；FFT；检测 中图分类号：TP216 文献标志码：A Detection of STN-LCD Defects Based on Image Registration SHEN Hui-liang，ZHANG Hong-gang ，LI Zhi-neng ( Department of Information and Electronic Engineering, Zhejiang University, Hangzhou 310027, China ) Abstract: a method based on the technique of image registration is proposed for detecting the appearance defects of Super-twisted Nematic Liquid Crystal Display (STN-LCD). The standard template LCD image is first recovered from possible non-uniform illumination, and then the statistical information of each region is collected by image thresholding and region labeling. The template and testing image are registered by control-point detection and affine transform. The defects including stroke-loss and pinhole of the LCD can then be detected by using region statistics such as mean and standard deviation of grayscales. To improve the accuracy of image registration, a special scheme of control-point selection and hybrid image interpolation is proposed. Experimental results show that the proposed method is robust and accurate, and can detect STN-LCD defects auto