基于BP神经网络的线Mura缺陷识别与定位研究

李一能,曾庆化,张月圆,等. 基于BP神经网络的线Mura缺陷识别与定位研究[J]. 光电工程,2020,47(11):190725. doi: 10.12086/oee.2020.190725
引用本文: 李一能,曾庆化,张月圆,等. 基于BP神经网络的线Mura缺陷识别与定位研究[J]. 光电工程,2020,47(11):190725. doi: 10.12086/oee.2020.190725
Li Y N, Zeng Q H, Zhang Y Y, et al. Mura detection and positioning in picture based on BP neural network[J]. Opto-Electron Eng, 2020, 47(11): 190725. doi: 10.12086/oee.2020.190725
Citation: Li Y N, Zeng Q H, Zhang Y Y, et al. Mura detection and positioning in picture based on BP neural network[J]. Opto-Electron Eng, 2020, 47(11): 190725. doi: 10.12086/oee.2020.190725

基于BP神经网络的线Mura缺陷识别与定位研究

  • 基金项目:
    国家自然科学基金资助项目(61533008, 61374115, 61603181);中央高校基本科研业务费专项资金(NJ20170005, NJ20170010);江苏高校优势学科建设工程
详细信息
    作者简介:
    通讯作者: 曾庆化(1979-),男,教授,主要从事视觉导航、卫星导航及多源信息数据融合导航的研究。E-mail:zengqh@nuaa.edu.cn
  • 中图分类号: TP391

Mura detection and positioning in picture based on BP neural network

  • Fund Project: Supported by the National Natural Science Foundation of China (61533008, 61374115, 61603181), the Fundamental Research Funds for the Central Universities (NJ20170005, NJ20170010), and the Priority Academic Program Development of Jiangsu Higher Education Institutions
More Information
  • 各类显示屏中Mura缺陷的自动识别和定位对提高显示屏幕的产品品质具有重要作用,是当前迫切需要发展的重要技术之一。针对当前手机屏幕Mura缺陷对比度低、缺乏明显边缘等特点,文中提出一种基于图像灰度曲线的Mura缺陷检测方法及其改进方法。改进方法基于均值滤波平滑和降采样原理,通过研究采样线上灰度曲线的波峰与波谷信息,利用BP神经网络构建线Mura缺陷的自动检测和定位算法。结合真实手机屏幕图像验证结果表明,与现有多种Mura缺陷检测方法相比,本文的改进方法能更准确快速地识别手机屏幕中的线Mura缺陷,识别准确率达到98.33%,检测过程无需调节参数,实现了线Mura缺陷的自动检测和定位。

  • Overview: Since the screen of the mobile phone has complex structures, and the manufacturing process is complicated, point Mura, and line Mura tend to arise when a screen is produced. Mura defect is a kind of display defect on the liquid crystal display (LCD) screen. It has various forms, such as uneven brightness or color in some areas of the screen, low contrast between the defect area and the surrounding background, blurred edges, and so on. These characteristics make the traditional methods based on edge detection and threshold segmentation difficult to detect Mura defect effectively.

    The traditional detection of Mura mainly depends on manual detection. So, this method has high labor cost, and the result of Mura detection is greatly influenced by the experience of workers. It clearly cannot meet the requirements of large batch orders of screen detection. Therefore, it is urgent to study the technology of automatic detection and location of Mura defect on the phone screen based on the computer. In recent years, many kinds of automatic detection methods are applied to detecting Mura defect. Due to the visibility of Mura defect on different display screens is different and various, a lot of parameters need to be adjusted from time to time during the whole detection process, leading relevant method wasting too much time if the number of screens is large.

    This paper proposes a new method of detecting line Mura based on gray curve of the image and BP neural network. Firstly, the image is preprocessed to reduce the influence of noise. Then the gray curves on the sampling lines in the image are analyzed to find out the location of the gray discontinuous parts. Since Mura defect often appears in the discontinuous parts of the gray sampling line of the image, it can help judge whether Mura is present on the screen.

    On this foundation, considering BP neural network is strongly nonlinear and with good robustness, the feature information reflecting whether Mura exists is extracted from the gray curves of the image for training. After training, the BP neural network can detect Mura defect automatically.

    The experimental results show that, compared with the existing Mura detection methods, the improved method in this paper can distinguish line Mura defect on the mobile phone screen more accurately and quickly. The accuracy rate is up to 98.33%, and no parameter needs to be adjusted during the detection process, realizing automatic detection and positioning of line Mura.

  • 加载中
  • 图 1  手机屏幕样本与缺陷所在位置。(a)横向Mura;(b)纵向Mura

    Figure 1.  Mobile screen samples and locations of Mura defect. (a) Horizontal Mura; (b) Vertical Mura

    图 2  图像降采样前后采样线灰度曲线示意图。(a)降采样后采样线灰度曲线;(b)未降采样采样线灰度曲线

    Figure 2.  Gray curves of sampling line before and after downsampling. (a) Gray curves of sampling line after downsampling; (b) Gray curves of sampling line before downsampling

    图 3  手机屏幕暗条纹线Mura对应关系示意图。(a)手机屏幕采样线示意图;(b)手机屏幕采样线灰度变化示意图

    Figure 3.  Connection between black line Mura and mobile screen. (a) Sampling lines on mobile screen; (b) Change of sampling lines on mobile screen

    图 4  检测孤点、极值点示意图

    Figure 4.  Locations of Isolated points and extreme points

    图 5  线Mura检测流程图

    Figure 5.  Flowchart of detecting line Mura

    图 6  BP 神经网输入值示意图

    Figure 6.  Input value of BP neural network

    图 7  BP网络的训练数据均方误差曲线图

    Figure 7.  Mean square error curve of BP network

    图 8  部分手机屏幕原图像。(a)本文算法未成功检测的屏幕图像;(b)人工漏检屏幕图像;(c)本文算法误检图像1;(d)本文算法误检图像2

    Figure 8.  Some original screen images. (a) The screen image not successfully detected by the algorithm proposed; (b) The screen image missed manually; (c) False check image 1; (d) False check image 2

    图 9  不同算法实验结果对比图。(a)原始图像(已增强对比度);(b)本文方法检测结果;(c)文献[9]方法检测效果;(d)文献[10]方法检测结果

    Figure 9.  Comparison of results of different algorithms. (a) Original image (enhanced contrast); (b) Detection results of the proposed method; (c) Detection results of Ref. [9]; (d) Detection results of Ref. [10]

    表 1  线Mura检测率

    Table 1.  Detection rate of line Mura

    序号 类别 样本个数 成功检测的样本数 检测成功率/%
    1 线 Mura 120 119 99.17
    2 无 Mura 60 53 88.33
    下载: 导出CSV

    表 2  不同算法对测试数据集中180幅图像进行检测的统计结果比较

    Table 2.  Comparison of results of 180 images in the test data detected by different algorithms

    方法 正确检测率/% 平均检测时间/s 检测过程是否需要调节参数
    灰度曲线法[1] 90.56 6.57 需要
    文献[9]方法 92.78 2.25 不需要
    文献[10]方法 43.89 7.32 需要
    (本文改进方法)基于图像灰度与BP神经网络方法 98.33 0.63 不需要
    下载: 导出CSV
  • [1]

    张鹏, 马婷婷, 杨叶花, 等.液晶显示器Mura缺陷及测量方法浅析[J].电子测试, 2017(6): 50-52. http://www.cqvip.com/QK/92717A/20173X/87689067504849554854485051.html

    Zhang P, Ma T T, Yang Y H, et al. Mura defect and measurement method of liquid crystal display[J]. Electronic Test, 2017(6): 50-52. http://www.cqvip.com/QK/92717A/20173X/87689067504849554854485051.html

    [2]

    Yang Y B, Li N, Zhang Y. Automatic TFT-LCD Mura detection based on image reconstruction and processing[C]//IEEE Third International Conference on Consumer Electronics, Berlin, 2014: 240-244.

    [3]

    Fan S K S, Chuang Y C. Automatic detection of Mura defect in TFT-LCD based on regression diagnostics[J]. Pattern Recognition Letters, 2010, 31(15): 2397-2404. doi: 10.1016/j.patrec.2010.07.013

    [4]

    石光明.基于亮度梯度的LCD屏的Black Mura缺陷检测[J].电视技术, 2018, 42(11): 40-43. http://www.cnki.com.cn/Article/CJFDTotal-DSSS201811009.htm

    Shi G M. Luminance gradient based LCD black Mura defect detection[J]. Video Engineering, 2018, 42(11): 40-43. http://www.cnki.com.cn/Article/CJFDTotal-DSSS201811009.htm

    [5]

    牟晨.基于图像处理的TFT-LCD Mura缺陷检测算法研究[D].长春: 吉林大学, 2018.

    Mu C. Research on TFT-LCD Mura defect detection algorithm based in image processing[D]. Changchun: Jilin University, 2018.

    [6]

    Ma Z Q, Gong J. An automatic detection method of Mura defects for liquid crystal display[C]// 2019 Chinese Control Conference (CCC), Guangzhou, China, 2019: 7722-7727.

    [7]

    Kong L F, Shen J, Hu Z L, et al. Detection of water-stains defects in TFT-LCD based on machine vision[C]//2018 11th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI), Beijing, China, 2018: 1-5.

    [8]

    Wang X, Dong R, Li B. TFT-LCD Mura defect detection based on ICA and multi-channels fusion[C]// 2016 3rd International Conference on Information Science and Control Engineering (ICISCE), Beijing, China, 2016: 687-691.

    [9]

    廖苗, 刘毅志, 欧阳军林, 等.基于自适应局部增强的手机TFT-LCD屏Mura缺陷自动检测[J].液晶与显示, 2018, 33(6): 475-482. http://www.zhangqiaokeyan.com/academic-journal-cn_chinese-journal-liquid-crystals-displays_thesis/0201271516109.html

    Liao M, Liu Y Z, Ouyang J L, et al. Automatic detection of Mura defect in TFT-LCD mobile screen based on adaptive local enhancement[J]. Chinese Journal of Liquid Crystals and Displays, 2018, 33(6): 475-482. http://www.zhangqiaokeyan.com/academic-journal-cn_chinese-journal-liquid-crystals-displays_thesis/0201271516109.html

    [10]

    钱基德, 陈斌, 钱基业, 等.基于机器视觉的液晶屏Mura缺陷检测方法[J].计算机科学, 2018, 45(6): 296-300, 313.

    Qian J D, Chen B, Qian J Y, et al. Machine vision based inspection method of Mura defect for LCD[J]. Computer Science, 2018, 45(6): 296-300, 313.

    [11]

    郝仕嘉, 周嘉琪.基于机器视觉的手机屏幕缺陷检测方法研究[J].信息与电脑, 2018(9): 50-53. http://www.cqvip.com/QK/95974A/20189/675293302.html

    Hao S J, Zhou J Q. Research on detecting method of mobile phone screen defect based on machine vision[J]. China Computer & Communication, 2018(9): 50-53. http://www.cqvip.com/QK/95974A/20189/675293302.html

    [12]

    赵晓雷.基于小波变换和均值滤波的图像去噪研究[J].信息技术, 2017(2): 69-71, 75. http://www.cnki.com.cn/Article/CJFDTOTAL-HDZJ201702016.htm

    Zhao X L. Image denoising based on wavelet transform and average filtering[J]. Information Technology, 2017(2): 69-71, 75. http://www.cnki.com.cn/Article/CJFDTOTAL-HDZJ201702016.htm

    [13]

    谷元保, 余康为, 姚洪涛.基于降采样块匹配的数字视频3D降噪算法[J].长春理工大学学报(自然科学版), 2016, 39(2): 98-102. http://d.wanfangdata.com.cn/Periodical/ccgxjmjxxyxb201602021

    Gu Y B, Yu K W, Yao H T. A digital video 3D denoising algorithm based on down sampled block matching[J]. Journal of Changchun University of Science and Technology (Natural Science Edition), 2016, 39(2): 98-102. http://d.wanfangdata.com.cn/Periodical/ccgxjmjxxyxb201602021

    [14]

    Cen Y G, Zhao R Z, Cen L H, et al. Defect inspection for TFT-LCD images based on the low-rank matrix reconstruction[J]. Neurocomputing, 2015, 149: 1206-1215. doi: 10.1016/j.neucom.2014.09.007

    [15]

    Han X H, Xiong X Y, Duan F. A new method for image segmentation based on BP neural network and gravitational search algorithm enhanced by cat chaotic mapping[J]. Applied Intelligence, 2015, 43(4): 855-873. doi: 10.1007/s10489-015-0679-5

    [16]

    杨守建, 陈恳. BP神经网络性能与隐藏层结构的相关性探究[J].宁波大学学报(理工版), 2013, 26(1): 48-52. http://www.cnki.com.cn/Article/CJFDTotal-NBDZ201301009.htm

    Yang S J, Chen K. Effect of multi-hidden-layer on performance of BP neural network[J]. Journal of Ningbo University (NSEE), 2013, 26(1): 48-52. http://www.cnki.com.cn/Article/CJFDTotal-NBDZ201301009.htm

  • 加载中

(9)

(2)

计量
  • 文章访问数:  4716
  • PDF下载数:  1263
  • 施引文献:  0
出版历程
收稿日期:  2019-12-11
修回日期:  2020-06-28
刊出日期:  2020-11-15

目录

/

返回文章
返回