TFT-LCD点线缺陷检测图像处理算法研究

发布时间：2018-08-23 19:50

【摘要】：薄膜晶体管液晶显示器(TFT-LCD)产业发展迅速,目前已经成功取代了CRT显示器。其中对面板进行缺陷检测是TFT-LCD生产线上重要的一部分,由于TFT-LCD面板结构复杂,生产流程多,要求无尘的生产环境等各种因素,不可避免的会产生各种缺陷,在各种缺陷中点线缺陷所占的比例最高,所以研究对点线缺陷的检测技术具有重要的意义。论文在研究TFT-LCD面板结构及显示原理的基础上,针对TFT-LCD面板表面图像具有重复性的纹理背景的特点,首先研究了抑制纹理背景的算法,通过抑制纹理背景进行重构图像,最后研究对重构图中缺陷区域分割的算法和提取缺陷特征的算法,并对各种算法进行实验分析。主要对以下几点进行了研究:(1)研究了TFT-LCD面板的结构、生产流程和显示原理,对点线缺陷的产生原因进行分析,并对点线缺陷进行了分类,最后列出了常见的点线缺陷,为下面进行缺陷检测算法研究打下基础。(2)通过研究缺陷检测图像处理的算法,并针对点线缺陷的特点,主要对以下两种算法进行了研究:一是在频域中处理的二维傅里叶变换法;二是在时域中处理的奇异值分解法,最后对两种算法进行对比分析并实验验证。(3)提出了结合双边滤波的加权模板差影检测算法。提出的加权模板差影法处理可以消除亮度不均匀的影响,在实际生产中由于相机对焦等问题进行差影处理后可能会有残余的纹理背景点,针对此问题采用双边滤波进行去除。对纹理背景抑制后图像采用最大熵法分割缺陷区域,最后对分割的缺陷区域的特点进行特征提取。(4)用所提算法对不同的点线缺陷样本图进行实验,从实验效果、处理时间和对亮度不均匀是否敏感方面进行比较分析,验证了本文提出的算法能有效地检测出缺陷并提取出相关参数。
[Abstract]:Thin film transistor liquid crystal display (TFT-LCD) industry is developing rapidly, and it has successfully replaced CRT display. Defect detection of TFT-LCD panel is an important part of TFT-LCD production line. In this paper, based on the study of the structure and display principle of TFT-LCD panel, the algorithm of restraining texture background is studied firstly, aiming at the feature that the surface image of TFT-LCD panel has repetitive texture background. Restraining texture background to reconstruct the image, the algorithm of defect region segmentation and defect feature extraction in the reconstructed image are studied, and various algorithms are analyzed experimentally. The main research points are as follows: (1) The structure, production process and display principle of TFT-LCD panel are studied, and the causes of dot-line defects are analyzed. This paper analyzes and classifies the point and line defects, and lists the common point and line defects, which lays a foundation for the following defect detection algorithm research. (2) By studying the algorithm of defect detection image processing, and aiming at the characteristics of point and line defects, the following two algorithms are mainly studied: one is the two-dimensional Fourier transform processed in the frequency domain. (3) A weighted template difference detection algorithm combined with bilateral filtering is proposed. The proposed weighted template difference method can eliminate the influence of uneven brightness and make difference in actual production due to camera focusing and other problems. After the texture background is suppressed, the defect area is segmented by maximum entropy method. Finally, the feature of the defect area is extracted. (4) Experiments are carried out on different defect sample images with the proposed algorithm. The processing time and the sensitivity to the non-uniformity of brightness are compared and analyzed. The results show that the proposed algorithm can effectively detect the defects and extract the relevant parameters.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN873.93;TP391.41

【参考文献】