TFT-LCD的Mura缺陷检测技术研究

发布时间：2018-04-26 10:37

本文选题：TFT-LCD + B样条曲面拟合　；参考：《合肥工业大学》2017年硕士论文

【摘要】：薄膜晶体管液晶显示器(TFT-LCD)在技术和生产方面发展迅速,正逐渐往轻薄化、大尺寸化、高分辨率化发展。在大尺寸和大规模生产的同时,LCD面板出现各类缺陷的几率也大大增加,不仅限制了产量,还提高了成本。针对TFT-LCD面板上Mura缺陷对比度低、边缘模糊、形状不规则的特点,在对大量检测方法的尝试后,论文重点研究并设计了一种在背景亮度不均匀条件下的TFT-LCD Mura缺陷检测技术。根据实际的检测需要,确立了高斯滤波去噪声、双三次B样条曲面拟合去除背景、最大类间方差(Otsu)的双γ分段指数变换对比度增强、Otsu法阈值分割缺陷和SEMU标准量化评价的检测流程,对检测过程中各阶段关键技术进行了分析和推导,并用Matlab软件编译实现。主要对以下几点进行了研究:(1)首先分析总结了Mura的定义、产生的原因和分类,然后归纳了图像噪声的来源和常用的四种去噪方法。(2)在曲面拟合去除背景部分,先总结了二元三次多项式曲面拟合和双三次B样条曲面拟合的理论知识,采用乘积型方程反算双三次B样条曲面并对拟合数据进行压缩以提高算法的速度和计算效率;针对传统的双三次B样条拟合精度过高的问题,引入光顺项对拟合曲面进行光顺平滑,以减小Mura区域点对拟合曲面局部形状的干扰。(3)介绍了两种已有的对比度增强方法并分析其缺点,从而提出一种新的增强方法:通过引入Otsu法自动选取阈值作为分段函数的分段点;引入γ指数变换,对背景和目标区域分别做不同的γ变换。所提方法能实现增强目标区域对比度并抑制背景区域灰度值变化,同时增强了Mura缺陷的边缘。再采用Otsu法阈值分割将图像二值化,并提取缺陷面积等相关参数。(4)引入SEMU标准,将前几阶段中提取的相关参数带入公式,求得用于评级的Semu值,以判定算法的准确性。将对比度增强和缺陷分割结合起来用以展示各算法的效果,证明了所提算法的实用性和高效性。通过Mura缺陷检测实验,总结归纳出检测算法流程。实验表明,本文所提检测算法对各类常见的Mura缺陷均能有效检出;对100个样本图像,在相关参数的设定下,有效检出率达99%。
[Abstract]:TFT-LCD (thin Film Transistor liquid Crystal display) is developing rapidly in technology and production, and is developing to thinning, large scale and high resolution. At the same time as large size and mass production, LCD panels have increased the probability of various defects, not only limiting production, but also increasing the cost. In view of the characteristics of low contrast, fuzzy edge and irregular shape of Mura defects on TFT-LCD panel, this paper focuses on the research and design of a TFT-LCD Mura defect detection technology under the condition of uneven background brightness. According to the actual needs of detection, Gao Si filter noise removal, double cubic B-spline surface fitting to remove the background, The detection flow of threshold segmentation defect and SEMU standard quantitative evaluation of double 纬 -segment exponential transformation of maximum inter-class variance is analyzed and deduced. The key techniques in each stage of the detection are analyzed and deduced, and implemented by Matlab software. Firstly, the definition, causes and classification of Mura are analyzed and summarized, then the source of image noise and four common denoising methods. Firstly, the theoretical knowledge of biquadratic cubic polynomial surface fitting and bicubic B-spline surface fitting is summarized. The product type equation is used to inverse calculate the bicubic B-spline surface and the fitting data are compressed to improve the speed and computational efficiency of the algorithm. Aiming at the problem of high precision of traditional bicubic B-spline fitting, the fairing term is introduced to smooth the fitting surface. Two existing contrast enhancement methods are introduced and their shortcomings are analyzed in order to reduce the interference of Mura region points to the local shape of fitting surface. A new enhancement method is proposed: the threshold value is automatically selected as the segmentation point of the piecewise function by introducing the Otsu method, and the 纬 -exponential transformation is introduced to make different 纬 transformations for the background and the target region respectively. The proposed method can enhance the contrast of the target area and suppress the change of the gray value of the background area, and enhance the edge of the Mura defect at the same time. Then the binarization of the image is made by using Otsu threshold segmentation method, and the relevant parameters such as defect area are extracted into the SEMU standard. The relevant parameters extracted in the previous stages are brought into the formula to obtain the Semu value used for rating, so as to determine the accuracy of the algorithm. Contrast enhancement and defect segmentation are combined to show the effectiveness of each algorithm, which proves the practicability and efficiency of the proposed algorithm. Through the Mura defect detection experiment, summed up the detection algorithm flow. The experimental results show that the proposed detection algorithm can effectively detect all kinds of common Mura defects, and the effective detection rate of 100 sample images can reach 99.9% under the setting of related 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】