【作者】 李茂；

【导师】 李辉；

【作者基本信息】 电子科技大学 ， 模式识别与智能系统， 2013， 硕士

【摘要】 随着液晶显示驱动技术的飞速发展，以薄膜晶体管液晶显示器（TFT-LCD）为代表的新型平板显示器逐渐成为了显示终端的主流。TFT-LCD由于具备低功耗、轻薄易用、高亮度、高对比度和高响应速度等众多优点，正受到越来越多的消费者的青睐，应用领域也在不断扩展。与此同时，为适应市场需求，TFT-LCD正向着高分辨率、轻薄化、低功耗和大屏幕的方向发展。在其生产过程中，TFT-LCD的重要组成部分玻璃面板等材料尺寸在逐步扩大，厚度在渐渐变薄，但随之出现的显示不均匀缺陷（Mura缺陷）问题也在日益突出。目前在液晶面板行业，绝大多数生产制造商在Mura缺陷检测环节仍然采用传统的人工视觉检测（HVI）方法，但该方法易受检测人员主观因素及外界环境影响，且对Mura缺陷缺乏客观的量化标准，很难保证产品质量，检测效率也极为低下，因此迫切需要研究稳定高效的Mura缺陷自动检测方法。本文在分析Mura缺陷典型特点的基础上，提出了一种基于机器视觉的Mura缺陷检测方法，该方法包含样本采集、预处理、背景抑制、亮度校正、Mura缺陷分割和Mura缺陷量化六大步骤。同时，本文综合利用相关图像处理理论，通过详细的原理推导、算法设计和实验仿真，分别实现了检测方法中背景抑制、亮度校正、Mura缺陷分割及量化三大关键技术的设计与验证，并最终完成了基于机器视觉的TFT-LCD屏Mura缺陷标准检测流程的建立。针对含Mura缺陷的图像背景抑制问题，本文采用了基于Gabor滤波的算法思想，并通过合理配置滤波器的各项滤波参数，成功抑制掉了图像中存在的纹理，同时由于Gabor的滤波特性使得检测结果更符合人眼判别标准。针对图像中存在的亮度不均匀现象，本文采用了基于盲源分离模型的图像亮度校正方法，将图像中存在的亮度不均匀以及莫尔条纹等影响Mura缺陷检测的众多因素统一当成图像亮度噪声进行抑制，利用同态变换和FastICA算法相结合的方式能够在先验知识未知的情况下成功完成图像的亮度校正。针对低对比度、边缘模糊的Mura缺陷分割及量化问题，本文采用了基于Chan-Vese主动轮廓模型的图像分割方法，并结合水平集理论实现了Mura缺陷目标轮廓的准确定位，同时根据SEMU标准，提取Mura缺陷区域的面积及对比度参数实现了Mura缺陷的客观量化。最后，本文利用提出的Mura缺陷机器视觉检测方法，从随机抽取的含Mura缺陷的50个TFT-LCD样本中成功检测出48个，检测成功率达到96%。测试结果表明所提出的Mura缺陷检测方法实际可行，满足设计需求且成功率较高。更多还原

【Abstract】 With the rapid development of LCD technology, a new type of flat panel displaynamed TFT-LCD is becoming the preference of display terminal gradually. TFT-LCDis now getting more and more ubiquitous due to its unique characteristics such as lowpower consumption, ultra-thin, high brightness and contrast, rapid response and so on.Meanwhile, to meet the market’s demand, the development trends of TFT-LCD will beoriented to high-resolution, thin, light, low power consumption and large screen. Duringits manufacturing process, the size of the glass substrate materials and the associatedcomponents will be expanded gradually. However, its thickness will be reduced, whichleads to a rapid increase of the probability which Mura defects occur. Most LCDmanufacturers still apply the traditional human visual inspection method to Muradefects detection. However, it’s not only interfered by uncertain factors from testingengineers and testing environment seriously, but also lack of objective quantitativecriteria. Hence, the detection method applied to Mura now is extremely inefficient anddifficult to guarantee the quality of service, which leads to an urgent need to study thestable and efficient automatic detection method of Mura defects.A Mura defect detection method based on machine vision, which consists ofimage acquisition, image preprocessing, background suppression, uneven luminanceadjustment, image segmentation and defect quantization, is proposed in the dissertationaccording to the analysis of the typical features of Mura defects. Simultaneously, thedissertation achieves key solutions to the three critical problems of backgroundsuppression, uneven luminance adjustment and image segmentation through detailedtheoretical derivation, algorithm design and simulation, and finally accomplishes theestablishment of detecting procedure of Mura defects in TFT-LCD based on machinevision. To solve the problem of background suppression, an algorithm based on2D-Gabor filter is proposed and successfully suppresses the textures in image throughconfiguring the filter parameters appropriately, which makes the detection resultscloser to the criterion of the human eyes. To adjust the uneven luminance existing in the images of the samples, a proposal based on blind source separation is put forward,which regards uneven luminance and moire fringe as multiplicative noise andsuppresses it through the combination of the homomorphic transform and FastICAalgorithm without prior knowledge. To fulfill the segmentation of low contrast Muradefect, a scheme based on chan-vese model is performed with the help of level setmethod, which reaches the actually profile of Mura defect. At the same time, the areaand contrast of Mura defect are used in the quantization according to SEMU.Finally, experimental detection is done and48out of50TFT-LCD samples withMura defects are successfully detected. The experimental result shows that thedetection method proposed in the dissertation is practical feasible and manages to fulfillthe required.更多还原