本研究致力於發展一套LCD彩色濾光片的自動瑕疵分類系統，藉由此系統來輔助人工做瑕疵分類。透過瑕疵的分類可以分析製程中那一階段有缺失，以維護生產線無瑕運作的目的。
我們的研究是在彩色濾光片製程中分類七種常見的瑕疵，這些瑕疵包含漏光色剝落、纖維異物、膜面線刮、金屬異物、透明異物、黑色異物、及光阻異物。我們同時取得瑕疵區塊的反射光及透射光彩色影像。瑕疵影像大小固定為640?480像素，並且不考慮取像光源不均的問題。對於同一個瑕疵，我們分別對反射光及透射光影像做瑕疵擷取，再結合這二種影像的擷取結果以分類瑕疵。反射光影像的擷取包含三個步驟：首先，我們將影像轉為灰階影像，再利用Otsu 自動取得門檻值將影像二值化。接著將二值化影像做水平投影，並去除水平累積量超過 400 像素的水平線，再利用形態學的閉合 (closing) 與連通物件標籤法 (connected component labeling algorithm) 保留瑕疵。在透射光影像中，我們判斷每個像素的顏色，並將顏色區分為紅、綠、藍、白、黑、與其它顏色，再分別對紅、綠、藍的像素做水平及垂直投影，藉此定位出透射光影像中的每個色點 (cell)。最後再將不屬於該色點顏色的像素標記出來，以便得到最完整的瑕疵。
擷取出完整的瑕疵後，我們分別對反射光及透射光影像中的瑕疵利用四種特徵分類瑕疵。四種由反射光影像取得的特徵依序是灰階變異數、緊實性 (compactness)、瑕疵區塊的長短軸比、及瑕疵面積佔其圈選的矩形面積 (bounding box) 比。四種由透射光影像取得的特徵依序是透明率、不透明率、混色率、及透白光率。
在分類步驟中，首先，我們利用透白光率來分辨漏光色剝落。再利用緊實性、瑕疵的長短軸比、及瑕疵面積佔其圈選的矩形面積比，將瑕疵區分為條狀及塊狀瑕疵。對於條狀瑕疵，我們利用灰階變異數來辨別纖維異物與膜面線刮。對於塊狀瑕疵，我們利用灰階變異數來區分塊狀瑕疵的紋理為鬆散型紋理瑕疵與緊實型紋理瑕疵。接著分別對這二種紋理的瑕疵，我們利用透明率、不透明率、及混色率將鬆散型紋理瑕疵分成出金屬異物與透明異物；將緊實型紋理的瑕疵分成光阻異物及黑色異物。
由上述做法及特徵對於檢測的七種瑕疵盡可能的保留完整的外形，對每種瑕疵的平均判定正確率可達80%。其中容易產生誤判的情況如膜面線刮因瑕疵擷取不完整，使得計算瑕疵特徵錯誤，而將該瑕疵誤判為漏光色剝落。另一種常見的誤判情形是光阻異物與黑色異物的誤判。由於光阻異物與黑色異物的外顯極為相似，因此不易區分。

In liquid crystal display (LCD) penal manufacturing, certain repairable panel defects can be found through defect classifying. However, most inspection equipments cannot classify defect, and the defects are classified by human presently. In order to promote the yield rate, it is important for automatic defect classification on production line. Real-time automatic defect classification also maintains production procedure through analyzing the defect. The flaw of production procedure can be found as soon as possible.
In this study, we focus on defects with the same field of view (FOV) of reflex-lighted and back-lighted image which size is 640480 pixels. The propose method consists of three stages. First, we use an automatic thresholding method, morphology, and connected component labeling algorithm to extract the defect from the reflex-lighted image, and we mark pixels which color is different to the cells in the back-lighted image. Then, we combine the results from reflex-lighted and back-lighted image detection. Second, we categorize the defect with shape into two categories: bar and massy defects. We use principal axis ratio, compactness, and bounded defect area to describe the shapes. For bar defects, we use gray-level variance of the defect to recognize fiber matter and dot scratch. For massy defects, defects are classified into incoherent and compact texture with gray- level variance of the defect. For incoherent defects, we recognize metal and transparent matters with opacity ratio. For compact defects, we use the same features of incoherent defect classification to recognize black and resist matters. At last, we utilize the features and propose a dichotomy classification procedure to classify the defect types, and the average accuracy of defect recognition is more than 80%.

[1] Castleman, K. R., Digital Image Processing, Prentice Hall, New Jersey, 1979.
[2] Drimbarean, A. and P. F. Whelan, "Experiments in colour texture analysis," Pattern Recognition Letters, vol.22, no.10, pp.1161-1167, 2001.
[3] Haralick, R. M., K. Shanmugam, and I. Dinstein, "Textural features for image classification," IEEE Trans. on Systems, Man and Cybernetics, vol.3, no.6, pp.610-621, 1973.
[4] Iivarinen, J. and A. Visa, "An adaptive texture and shape based defect classification," in Proc. 14th Int. Conf. Pattern Recognition, Brisbane, Australia, Aug.16-20, 1998, pp.117-122.
[5] Jain, R., R. Kasturi, and B. G. Schunck, Machine Vision, McGraw-Hill, New York, 1995.
[6] Kaplan, L. M., "Extended fractal analysis for texture classification and segmentation," IEEE Trans. Image Processing, vol.8, no.11, pp.1572-1585, 1999.
[7] Kim, Y. B. and J. Gao, "A new semi-supervised subspace slustering slgorithm on fitting mixture models," in Proc. IEEE Symp. on Computational Intelligence in Bioinformatics and Computational Biology, San Diego, CA, Nov.14-15, 2005, pp.1-8.
[8] Kirshner, S., I. V. Cadez, P. Smyth, and C. Kamath, "Learning to classify galaxy shapes using the EM algorithm," in Proc. 15th Neural Information Processing Systems, Vancouver, British Columbia, Canada, Dec.9-14, 2002, pp.1497-1504.
[9] Lin, Z., J. Wang, and K. K. Ma, "Using eigencolor normalization for illumination-invariant color object recognition," Pattern Recognition, vol.35, no.11, pp.2629-2642, 2002.
[10] Lin, H. D. and C. H. Chien, "Automated detection of color non-uniformity defects in TFT-LCD," in Proc. Int. Joint Conf. on Neural Networks, Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada, Jul.16-21, 2006, pp.2384-2391.
[11] Lyu, S., "A kernel between unordered sets of data: the Gaussian mixture approach," in Proc. 16th Euro. Conf. Machine Learning, Porto, Portugal, Oct.3-7, 2005, pp.255-267.
[12] Mahtab, A., V. N. Sridhar, and R. R. Navalgund, "Impact of surface anisotropy on classification accuracy of selected vegetation classes: an evaluation using multidate multiangular MISR data over parts of madhya pradesh, India," Geoscience and Remote Sensing, vol.46, no.1, pp.250-258, 2008.
[13] Malhi, A. and R. X. Gao, "Feature selection for defect classification in machine condition monitoring," in Proc. 20th IEEE Instrumentation Measurement Technology Conf., Massachusetts, May 20-22, 2003, pp.36-41.
[14] Malhi, A. and R.X. Gao, "PCA-based feature selection scheme for machine defect classification," IEEE Trans. Instrumentation and Measurement, vol.53, no.6, pp.1517-1525, 2004.
[15] Mallat, S. and S. Zhong, "Characterization of signals from multiscale edges," IEEE Trans. Image Pattern Anal. Machine Intell., vol.14, no.7, pp.710-732, 1992.
[16] Otsu, N., "A threshold selection method from gray-level histograms," IEEE Trans. System Man and Cybernetics, vol.9, no.1, pp.62-66, 1979.
[17] Petrou, M. and P. Bosdogianni, Image Processing The Fundamentals, John Wiley & Sons, New York, 1999.
[18] Rohrmus, D. R., "Invariant and adaptive geometrical texture features for defect detection and classification," Pattern Recognition, vol.38, no.10, pp.1546-1559, 2005.
[19] Shuttleworth, J. K., A. G. Todman, R. N. G. Naguib, B. M. Newman and M. K. Bennett, " Colour texture analysis using co-occurrence matrices for classification of colon cancer images," in Proc. IEEE Canadian Conf. on Electrical and Computer Engineering, Winnipeg, Canada, May 12-15, 2002, pp.1134-1139.
[20] Song, K. Y., J. Kittler and M. Petrou, "Defect detection in random colour textures," Image and Vision Computing, vol.14, no.9, pp.667-683, 1996.
[21] Song, Y. C., D. H. Choi, and K. H. Park, "Multiscal detection of defect in thin film transistor liquid crystal display panel," Japanese Journal of Applied Physics, vol.43, no.8A, pp.5465-5468, 2004.
[22] Teppei, I., T. Takateru, J. I. Hayashi, and S. Hata, "Visual defects classification system using co-occurrence histogram image," in Proc. Annual Conf. SPIE, Japan, Sep.17-20, 2007, pp.598-603.
[23] Tsai, D. M. and C. Y. Hung, "Automatic defect inspection of patterned thin film transistor-liquid crystal display (TFT-LCD) panels using one-dimensional fourier reconstruction and wavelet decomposition," Int. Journal of Production Reserach, vol.43, no.21, pp.4589-4607, 2005.
[24] Wang, J. and A. K. Asundi, "A computer vision system for wineglass defect inspection via Gabor-filter-based texture features," Information Sciences, vol.127, no.3, pp.157-171, 2000.
[25] Wang, H. X., B. Luo, Q. B. Zhang, and S. Wei, "Estimation for the number of components in a mixture model using stepwise split-and-merge EM algorithm," Pattern Recognition Letters, vol.25, pp.1799-1809, 2004.
[26] Xie, X. and M. Mirmehdi, "Texture exemplars for defect detection on random textures," in Proc. 3rd Int. Conf. Advances in Pattern Recognition, Bath, UK, Aug.22-25, 2005, pp.404-413.
[27] Xie, X. H., M. Mirmehdi, and B. Thomas, "Colour tonality inspection using eigenspace features," Machine Vision and Applications, vol.16, no.6, pp.364-373, 2006.
[28] Yang, X., G. Pang, and N. Yung, "Discriminative training approaches to fabric defect classification based on wavelet transform," Pattern Recognition, vol.37, no.5, pp.889-899, 2004.
[29] Yin, P. Y. and L. H. Chen, "A fast iterative scheme for multilevel thresholding methods," Signal Processing, vol.60, no.3, pp.305-313, 1997.
[30] Zhang, X., C. Krewet, and B. Kuhlenk?tter, "Automatic classification of defects on the product surface in grinding and polishing," International Journal of Machine Tools and Manufacture, vol.46, no.1, pp.59-69, 2006.
[31] http://en.wikipedia.org/wiki/Repeatability
Repeatability, from Wikipedia.