基于机器视觉的液晶屏点灯缺陷检测系统关键技术研究

发布时间：2018-05-22 07:40

本文选题：TFT-LCD + 陷波滤波　；参考：《湖南理工学院》2017年硕士论文

【摘要】：液晶屏的制造过程极其复杂,需要近百道工序,在制造的过程中难免会出现许多类型的缺陷。因此对液晶屏缺陷的检测在生产过程起着关键性作用。利用机器视觉技术实现液晶屏点灯缺陷的高速自动检测是液晶屏自动化生产的重要研究课题之一。论文通过分析液晶屏点灯图像中存在的缺陷和瑕疵的特征,提出基于图像处理的快速、有效的缺陷检测方法。论文主要进行了以下几方面的研究:子像素级缺陷检测方法研究:传统的图像学处理方法很难检测出子像素级缺陷。我们通过研究液晶屏图像的特点,发现同一型号液晶屏中像素均是规则排列,因此对缺陷图像经过傅里叶变换能得到一致的频谱图,针对此特点,论文提出了一种基于陷波滤波和图像配准的子像素缺陷检测方法。首先用无缺陷模板图像建立配准模板和陷波滤波模板;然后用配准模板对缺陷图像进行图像配准,解决屏幕偏移问题;再用陷波滤波模板进行滤波处理,滤除背景纹理,使缺陷更加明显;最后对图像阈值分割,找出缺陷。结果显示,该方法可以准确、快速的检测子像素级缺陷。中小型缺陷检测方法研究:由于相机拍摄的液晶屏图像缺陷、背景、细节、噪声等信息都包含在一个较窄的灰度范围内,导致缺陷与其他信息难以区分,因此,论文提出了一种融合局部熵与局部均匀度液晶屏缺陷检测方法。将像素分布的局部熵值和局部均匀度值相结合,滤除空间邻域内噪声并获得像素的空间分布,利用空间特征实现对中小型缺陷的检测。结果显示,论文方法不需要进行频域处理便可准确找到缺陷,检测速率较高,相比于其他算法,有较强鲁棒性。Mura缺陷检测方法研究:液晶屏Mura缺陷均具有背景整体亮度不均、灰度变化不明显等特点,用基于机器视觉的方法从中检测出缺陷是非常困难的。论文提出一种新的Mura缺陷检测方法,首先对液晶屏图像中背景与Mura缺陷灰度值差异进行分析,然后采用均值滤波和背景差分法来抑制背景杂波,再利用灰度约束获取疑似Mura缺陷区域,最后提取疑似区域的灰度特征放入训练后的BP神经网络提取缺陷目标。结果显示,论文方法有较好的除噪效果,检测率较高。
[Abstract]:The manufacturing process of LCD screen is extremely complex and requires nearly 100 processes. Many kinds of defects will inevitably appear in the manufacturing process. Therefore, the detection of LCD screen defects plays a key role in the production process. It is one of the important research topics in the automatic production of LCD screen to realize high speed automatic detection of the defects of LCD screen with machine vision technology. Based on the analysis of defects and defects in LCD screen lamp images, a fast and effective defect detection method based on image processing is proposed in this paper. This paper mainly studies the following aspects: subpixel level defect detection method: traditional image processing methods are difficult to detect sub-pixel level defects. By studying the characteristics of the LCD screen image, we find that the pixels in the same LCD screen are arranged regularly, so we can get the consistent spectrum of the defective image by Fourier transform. In this paper, a subpixel defect detection method based on notch filtering and image registration is proposed. First, the registration template and notch filter template are established by using the non-defect template image; then, the defect image is registered with the registration template to solve the screen offset problem; and then the notch filter template is used to filter the background texture. Make the defect more obvious; finally, the image threshold segmentation, find out the defect. The results show that the method can detect subpixel defects accurately and quickly. Research on small and medium defect Detection methods: because the defects, background, details, noise and other information of LCD screen images taken by camera are all contained in a narrow gray range, it is difficult to distinguish the defects from other information. In this paper, a method for defect detection of liquid crystal screen with fusion of local entropy and local uniformity is proposed. The local entropy value and local uniformity value of pixel distribution are combined to filter the noise in the spatial neighborhood and to obtain the spatial distribution of the pixels. The detection of small and medium-sized defects is realized by using spatial features. The results show that the method can accurately find the defects without frequency domain processing, and the detection rate is higher. Compared with other algorithms, the method of robust. Mura defect detection method: the Mura defects of LCD screen have the uneven brightness of the background as a whole. It is very difficult to detect defects by machine vision based method because the gray level change is not obvious. In this paper, a new Mura defect detection method is proposed. Firstly, the difference between background and Mura defect gray value in LCD screen image is analyzed, and then the mean filter and background difference method are used to suppress background clutter. Then the suspected Mura defect region is obtained by using gray constraints, and the gray feature of the suspected area is finally extracted into the trained BP neural network to extract the defect target. The results show that the method has better denoising effect and high detection rate.
【学位授予单位】：湖南理工学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN873.93;TP391.41

【参考文献】