隨著電子裝置普及，液晶顯示器的產量隨之上升，成為國內重要產業之一。大量生產中品質穩定度是工業非常重要的一環，各家廠商均需投入產品品質檢測。然而傳統的顯示器檢測仍大量倚賴人力，為了增加產品穩定度與競爭力，國際組織如SEMI於2002年提出人工檢測的指標，但後續研究指出人工檢測還有許多其他主觀因素影響檢測標準的穩定性。近年檢測導入AOI機台，但對於缺陷分類與品質評估仍靠人工判斷，因此急需發展更完善的檢測方式取代傳統檢測。電腦檢測有穩定、高效率的特性，研究方向主要分為缺陷分類與缺陷區提取兩大方向，但尚未有國際通用的電腦檢測指標。其中缺陷分類模型需要大量的影像進行模型訓練、而缺陷區提取方法則是有無法同時兼顧整體亮度異常偵測與局部缺陷偵測的問題。有鑑於此問題，本論文發展液晶顯示器的電腦檢測與評估方法，以卷積類神經網路結合影像處理實現缺陷分類方法，使用少量的影像完成高預測力模型訓練。透過人因工程實驗定量分析影響人工檢測的因素，並且透過迴歸方法找出顯著影響的因子。接著改善現有缺陷提取影像處理方法，加入雜訊處理、人眼閾值來改善提取面積不準的問題。最後以缺陷面積、缺陷梯度、缺陷平均亮度定義新的品質評估關係式，改善SEMI指標忽略梯度對人眼感受的問題、更接近人工檢測實驗結果。未來可以延續本研究方法完成更多分類的類神經模型與其他類型缺陷的程度評估。我們相信電腦檢測與評估方法不但可以使液晶顯示器檢測製程更加自動化且更穩定，未來相同的檢測模式可以套用到其他產品非破壞性檢測(NDT)上，配合分類、評估、成因分析，實現檢測製程的自動化與即時診斷。

With the increasing demands in modern mobile and consumer electronics, production and performance requirement of liquid crystal displays (LCD) is growing rapidly. In mass production, good and stable quality control is crucial, therefore, most manufacturers put extremely amount of efforts for quality inspection. However, up to now, LCD inspection process still highly depends on manually inspection human vision. To increase product stability and competitiveness, Semiconductor Equipment and Materials International (SEMI) has released quantized index for human inspection. Recently, automated optical inspection (AOI) was adopted to LCD inspection, but defect type classification and overall quality evaluation still relies on human inspectors. Previous studies showed that visual inspection is inevitably biased by the physical and psychological conditions and a more objective manner should be developed. Therefore, it is necessary to develop better automatic quality assurance methods based on computerized inspections such as image processing and defect classifications. Although there are many published researches focused on defect classification and defect extraction, there is still no internationally common index of computerlized inspection for LCD. In addition, those defect classification models also need large amount of images for training and the defect extraction process still faces problems in dealing with both overall brightness anomaly detection and local defect detection at the same time and this represents a block for preventing the application of image processing in LCD inspection. Therefore, the present thesis focuses on developing a method for addressing the above needs. Specifically, this thesis develops a rational flow to classify and to quantify the possible LCD defects by using convolutional neural network (CNN) and image processing to realize a high predictive defect classification scheme with only a minimal amount of training images. By conducting experiment and analyzing the relationship between human visual threshold and factors of defects, the dominant factors influencing human inspection can be identified. The next step is to enhance existing defect extraction image processing by artificially including the possible human factors such as adding of noise processing and human eye threshold. This allows us to simultaneously consider the characteristics of both computer and human inspections to further enhance the accuracy in defected area extraction. Finally, a novel quality evaluation index is proposed by integrating the information of defected area, defected gradient, and average brightness of defected zone, to improve the existed index of SEMI since they completely ignore the influence the perception of gradient of defects to human eyes. In summary, the experiment results of this thesis has shown that the application of the improved defect index produced better results close to human inspection. In the future, based on the methodology outlined in this thesis and include more types of defects in LCD, it is expected a more substantial defect classification and evaluation model suitable for industrial application can be finally realized. This would certainly benefit to LCD manufactures and the same methodology can also be adapted to other image-based nondestructive testing (NDT).

[1] 田民波、林怡欣, 平面顯示器之技術發展, 臺北市: 五南圖書出版公司, 2008
[2] 郭志忠, “TFT-LCD自動光學檢測-以Cell Zone為基礎的瑕疵分類研究,” 碩士論文, 國立成功大學, 臺南市, 2008
[3] 凌雲光子, ”LCD行業質量檢測解決方案,” 2005, [Online], Available: http://www.lustersys.com/d45b1ebc-7735-95ce-fc7a-86f9751586c5/4dbfc66a-2343-4dc5-813c-546de66911cf.shtml.[Accessed Jul. 7, 2019]
[4] 晶彩科技, ”檢測機台,” 2019, [Online], Available: https://www.favite.com/product/fine-resolution-array-aoi-1um/.[Accessed Jul. 7, 2019]
[5] Vizio TV Help, ”Line Defects,”2015, [Online], Available: https://vizhdtv.com/line-defects-on-vizio-tv-screens/.[Accessed Jul. 7, 2019]
[6] SEMI, DRAFT DOCUMENT 3324, 2002
[7] Y. Mori, R. Yoshitake, T. Tamura, T. Yoshizawa and S. Tsuji, “Evaluation and discrimination method of ‘mura’ in liquid crystal displays by just noticeable difference observation, ” Proc. of SPIE, vol.4908, pp.715-720, 2002
[8] 陳俊志,人眼閾值在液晶顯示器不均勻現象之探討, 碩士論文, 國立清華大學, 新竹市,2005
[9] Y. H. Won, H. Joo and J. S. Kim, “Classification of Defects in the Polarizer of Display Panels using the Convolution Neural Network (CNN), ” Int'l Journal of Computing, Communications & Instrumentation Engg., vol. 4, no. 1, pp.139-142, 2017
[10] H. Yang, S. Mei, K. Song, B. Tao and Z. Yin, "Transfer-Learning-Based Online Mura Defect Classification," in IEEE Transactions on Semiconductor Manufacturing, vol. 31, no. 1, pp. 116-123, Feb. 2018.
[11] J. Y. Lee and S. I. Yoo, “Automatic Detection of Region-Mura Defect in TFT-LCD,” IEICE TRANSACTIONS on Information and Systems, Vol.87-D, No.10, pp.2371-2378,2004
[12] J. S. Ryu, J. H. Oh, J. G. Kim, T. M. Koo and K. H. Park, “TFT-LCD panel Blob-Mura inspection using the correlation of wavelet coefficients,” 2004 IEEE Region 10 Conference TENCON 2004., Chiang Mai, pp. 219-222 Vol. 1, 2004
[13] L.C. Chen and C.C. Kuo, “Automatic TFT-LCD mura defect inspection using discrete cosine transform-based background filtering and ‘just noticeable difference' quantification strategies,” Measurement Science and Technology, vol. 19, No. 1, 2008
[14] K. Taniguchi, K. Uetaa and S. Tatsumi, “A mura detection method”, Pattern Recognition, vol. 39, no. 6, pp.1044-1052, 2006
[15] 張宇凱, 液晶顯示器可靠度性質與表現評估之方法建構, 碩士論文, 國立成功大學, 臺南市, 2017
[16] 張晉誠,中尺寸薄膜電晶體液晶顯示器因COG封裝所造成局部翹曲即應力對漏光現象之探討, 碩士論文, 國立交通大學, 新竹市, 2017
[17] 林泰印, 偏光板元件潮濕變形對液晶面板Mura之影響, 博士論文, 國立交通大學, 新竹市, 2009
[18] J. Leu, C. Chang, A. Chen, M. H. Lin and F. Huang, “Effects of localized warpage and stress on chip-on-glass packaging: Induced light-leakage phenomenon in mid-sized TFT-LCD,” Journal of the Society for Information Display, vol.20, no.1, pp.28-36, 2012.
[19] N. Otsu, “A Threshold Selection Method from Gray-Level Histograms,” in IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, Jan., pp. 62-66, 1979.
[20] M. Bellis, “The History of Liquid Crystal Display,”thoughtco.com, Mar. 2, 2019. [Online]. Available: https://www.thoughtco.com/liquid-crystal-display-history-lcd-%201992078. [Accessed Jul. 7, 2019]
[21] 田民波, TFT-LCD面板設計與構裝技術,臺北市:五南圖書出版公司,2008
[22] Y. Mohamed, “How do LCDs work?”, Apr. 21, 2015.[Online]. Available: http://alexsb.org/2015/04/21/how-do-lcds-work/ [Accessed Jul. 7, 2019].
[23] VESA, “330-x Mura Defects”, Jan. 2002
[24] B. Farell and D. G. Pelli, “Psychophysical methods, or how to measure a threshold and why,” in Vision Research: A Practical Guide to Laboratory Methods, R. H. S. Carpenter and J. G. Robson, Eds. New York: Oxford University Press, 1999, pp. 129-136
[25] Y. Le Cun, “Generalization and Network Design Strategies,” University of Toronto Connectionist Research Group, Toronto, Canada, Tech.Report. CRG-TR-89-4, Jun. 1989
[26] Y. Le Cun, Y. Bengio and G. Hilton, “Deep learning,” Nature, vol. 521, May, pp.436-444, 2015
[27] A. Krizhevsky, I. Sutskever and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” Advances in Neural Information Processing Systems 25, pp. 1097-1105, 2012
[28] K. Simonyan, A. Zisserman, “ Very Deep Convolutional Networks for Large-Scale Image Recognition,” arXiv:1409.1556v6 [cs.CV] 10, Apr. 2015
[29] 林大貴, tensorflow+Keras深度學習與人工智慧實務應用, 新北市:碩博出版社, 2017
[30] Y. Feunteun, “Parallel and Distributed Deep Learning: A Survey,” towardsdatascience.com, Apr. 30 , 2019. [Online]. Available:https://towardsdatascience.com/parallel-and-distributed-deep-learning-a-survey-97137ff94e4c.[Accessed Jul. 7, 2019]
[31] Y. Gan and Q. Zhao, “An Effective Defect Inspection Method for LCD Using Active Contour Model,” in IEEE Transactions on Instrumentation and Measurement, vol. 62, no. 9, Sep., pp. 2438-2445, 2013.
[32] G. Deng and L. W. Cahill, “An adaptive Gaussian filter for noise reduction and edge detection,” 1993 IEEE Conference Record Nuclear Science Symposium and Medical Imaging Conference, San Francisco, CA, USA, pp. 1615-1619 vol.3, 1993.
[33] C. Alard and R. H. Lupton, “A Method for Optimal Image Subtraction,” The Astrophysical Journal, vol. 503, no. 1, Aug., pp. 325-331, 1998.
[34] I. Sobel and G. Feldman, “A 3x3 Isotropic Gradient Operator for Image Processing,” in Pattern Classification and Scene Analysis, R. O. Duda and P. E. Hart, Eds. New York: John Wiley & Sons, pp. 271-272, 1973.
[35] Matlab Library
 
[36] R. O. Duda, P. E. Hart and D. G. Stork, Pattern Classification, New York: John Wiley & Sons, 2000.
[37] J. Canny, “A Computational Approach to Edge Detection,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-8, no. 6, pp. 679-698, Nov. 1986.
[38] LCDTech, “ LCD pixels structure,” May 14, 2012. [Online]. Available:http://lcdtech.info/en/tests/lcd.pixels.structure.htm. [Accessed Jul. 7, 2019]
[39] A.K.Jain, Fundamentals of Digital Image Processing, Upper Saddle River, NJ: Prentice Hall Press, 1989.
[40] A. Rosenfeld, and A. C. Kak, Digital Picture Processing, Cambridge, MA: Academic Press, 1982.
[41] R. C. Gonzalez and R. E. Woods, Digital Image Processing, London: Pearson, 2018.
[42] D. Padfield, “The magic sigma,” CVPR 2011, Colorado Springs, CO, USA, pp. 129-136, 2011.
[43] J. S. Weszka, R. N. Nagel and A. Rosenfeld, “A Threshold Selection Technique,” IEEE Transactions on Computers, vol. C-23, no. 12, Dec., pp. 1322-1326, 1974.
[44] R. Bogacz, E. Brown, J. Moehlis, P. Holmes, J. D. Cohen, “The physics of optimal decision making: A formal analysis of models of performance in two-alternative forced-choice tasks,” Psychological Review, vol. 113, no. 4, pp.700-765, 2006.
[45] R. H. Myers, Classical and modern regression with applications, Duxbury MA: Duxbury Thomson Learning,1986.
[46] M. Norusis, SPSS 16.0 Advanced Statistical Procedures Companion, Upper Saddle River, NJ: Prentice Hall Press, 2008.
[47] A. M. Brown, “A step-by-step guide to non-linear regression analysis of experimental data using a Microsoft Excel spreadsheet,” Computer Methods and Programs in Biomedicine, vol. 65, no. 3, pp.191-200, 2001.
 
[48] M. M. Yusr, S. S. M. Mokhtar, A. R. Othman and Y. Sulaiman, “Does interaction between TQM practices and knowledge management processes enhance the innovation performance?” International Journal of Quality & Reliability Management, vol. 34, no. 7, pp.955-974, 2017.
[49] Z. D. Wang, Y. Gu, Y. S. Wang, “A review of three magnetic NDT technologies, ” Journal of Magnetism and Magnetic Materials, Vol. 324, Issue 4, Feb., pp. 382-388, 2012.
[50] W. H. Yang and Y. S. Tarng, “Design optimization of cutting parameters for turning operations based on the Taguchi method,” Journal of Materials Processing Technology, Vol. 84, Issues 1–3, 1 Dec., pp. 122-129, 1998.