微結構滾輪壓印3D光學膜光柵時，其結構若有瑕疵，將對產線上3D光學膜光柵結構產生影響，因此微結構滾輪之瑕疵檢測，為3D光學膜製程成功與否之關鍵。本研究擬開發一套微結構滾輪瑕疵檢測系統，檢測刻痕缺陷、污漬、刮傷、粉塵與刻痕歪斜等目前產線上常產生之五種瑕疵。
微結構滾輪依據面板所需之尺寸與視點數，結構寬度與形狀亦會隨之改變，本研究利用其結構重複的特性，自動尋找無瑕疵模板。係藉由正交投影法將重複影像降低為一維資訊，搭配一維離散傅立葉轉換(Discrete fourier transform)，由頻率域求得影像週期，並藉影像週期推算重複影像大小，建立多張重複圖案之集合，再以基因演算法(Genetic algorithm)搭配影像資訊量之適應性函數，自動篩選此集合中無瑕疵影像，當作微結構滾輪之標準模板，接著利用影像相減搭配遲滯二值化(Hysteresis thresholding)，將瑕疵精確分割，擷取準確之特徵，最後將瑕疵特徵值輸入至羅吉斯模型樹分類器(Logistic model trees)訓練。本研究開發之微結構滾輪瑕疵檢測系統，整體辨識率達95.2%，並藉由檢測結果，回饋製程上並實際改善，亦經實務驗證於產線，確可提升生產良率及降低成本。

In the microstructure roller imprinting of 3D optical film gratings, if the structure has any defects, it would result in an adverse impact on the 3D optical film gratings structures on the production line. Therefore, the defect detection of the microstructure roller is the key to the success of the 3D optical film process. This study developed a microstructure roller defect detection system for the detection of five common defects including notch, stain, scratch, dust and nick skew on the current production line.
According to the dimensions and number of viewpoints required by the panel, the structural width and shape of the microstructure roller would change accordingly. This study utilized the characteristics of structural repetitiveness to automatically seek the flawless template. The orthogonal projection approach was applied to reduce the repeated images into one-dimensional information. Using the one-dimensional Discrete Fourier Transform, this study obtained the image cycle from the frequency field to deduce the size of the repeated image, and built the set of multiple repeated images. Next, by integrating GA (Genetic Algorithm) and the adaptability function of the image information quantity, the flawless images were automatically selected in the set as the standard templates for the microstructure roller. Next, by using image deduction coupled with Hysteresis Threshold, the defects were accurately segmented to extract the correct features. Finally, the defect features were inputted into the Logistic model trees for training. The overall detection rate of the proposed microstructure roller defect detection system reached 95.2%. The detection results were fed back to the process for actually improvement. The application in production line suggested that the system could improve product yield and reduce costs.

[1] Davies, E. R., “Machine Vision,” Academic Press, New York, pp. 211-244(1997).
[2] Schalkoff, R. J., “Digital Image Processing and Computer Vision,” John Wiley & Sons Inc, NY, pp. 279–286(1989).
[3] Mattoccia, S. Tombari F., and Di Stefano, L., “Fast Full-Search Equivalent Template Matching By Enhanced Bounded Correlation,” IEEE Trans on Image Processing; Vol. 17, No. 4, pp. 528-538(2008).
[4] Khalaj, B. H., Aghajan, H. K., and Kailath, T., “Digital Image Processing Techniques for Patterned Wafer Inspection,” SPIE Vol. 1926, pp. 508-1113(1993).
[5] Xie, P., and Guan S., “A Golden Template Self-generating Method for Patterned Wafer Inspection,” Machine Vision and Applications, Vol. 12, No. 3, pp. 149-156(2000).
[6] Kuo, C. F. J., and Shih, C. Y., “Printed Fabric Computerized Automatic Color Separating System,” Textile Research Journal, Vol. 81, No. 7, pp. 706–713(2010).
[7] 陳威仰，適應性相減法於週期性紋路之表面瑕疵檢測，元智大學，工業工程與管理學系，碩士論文(2008)。
[8] Otsu, N., “A Threshold Selection Method from Gray Level Histogram,” IEEE Trans. on Systems, Man, and Cybernetics, Vol. 9, No. 1, pp. 62-66(1979).
[9] Johannsen, G., and Bille, J., “A Threshold Selection Method Using Information Measures,” Proc. 6th Int. Conf. Pattern Recognition Munich, Germany, pp. 140-143(1982).
[10] Kapur, J. N., Sahoo, P. K., and Wong, A. K. C., “A New Method for Gray-Level Picture Thresholding Using the Entropy of The Histogram,” Computer Vision, Graphics, and Image Processing, Vol. 29, No. 3, pp. 273-285(1985).
[11] Cao, L., Shi, Z. K., and Cheng, E. K. W., “Fast Automatic Multilevel Thresholding Method,” Electronics Letters, Vol. 38, No. 16, pp. 868-870(2002).
[12] Ridler, T. W., and Calvard, S., “Picture Thresholding Using an Iterative Selection Method,” IEEE Transactions on Systems, Man and Cybernetics, Vol. 8, No. 8(1978).
[13] Tony P. P., “Thresholding Images of Line Drawings with Hysteresis,” Springer-Verlag London, UK, pp. 310-319(2002).
[14] Chen, F. L., and Liu, S. F., “A Neural-Network Approach To Recognize Defect Spatial Pattern In Semiconductor Fabrication,” IEEE Transactions on Semiconductor Manufacturing, Vol. 13, No. 3, pp. 366-373(2000).
[15] Tsai, D. M., and Yang, C. H. “A Quantile–Quantile Plot Based Pattern Matching for Defect Detection,” Pattern Recognition Letters, Vol. 26, No. 13, pp. 1986-1962(2005).
[16] Su, C. T., Yang, T., and Ke, C. M., “Neural-Network Approach for Semiconductor Wafer Post-Sawing Inspection,” IEEE Transactions on Semiconductor Manufacturing, Vol. 15, No. 2, pp. 260-266(2002).
[17] 黃怡菁，3D立體顯示技術，科學發展，451期，46-52頁(2010)。
[18] 鍾國亮，影像處理與電腦視覺，東華書局，台北(2008)。
[19] Gonzalez, R. C., and Woods, R. E., “Digital Image Processing,” Prentice Hall, Taipei(2008).
[20] Canny, J., “Finding Edges and Lines in Images,” M.I.T. Artificial Intelligence Lab, Cambridge(1983).
[21] Bouaynaya, N., “Theoretical Foundations of Spatially-Variant Mathematical Morphology Part II: Gray-Level Images,” IEEE Transactions On Pattern Analysis and Maxhine Intelligence, Vol. 30, No. 5, pp. 831-850(2008).
[22] Quinlan, J. R., “C4.5: Programs for Machine Learning, Morgan Kaufmann Publishers,” San Mateo, CA(1993).
[23] Quinlan, J. R., “Induction of Decision Tress,” Machine Learning, Vol. 1, No. 1, pp. 81-106(1986).
[24] Shannon, C. E., “A Mathematical Theory of Communication,” Bell System Technical Journal, Vol. 27, pp. 379-423(1948).
[25] Bratko, I., and Muggleton, S., “Applications of Inductive Logic Programming,” Communications of the ACM, Vol. 38, No. 11, pp. 65-70(1995).
[26] Han, J., and Kamber, M., “Data Mining: Concepts And Techniques,” Morgan Kaufmann, New York(2001).
[27] Russell, S., and Norvig, P., “Artificaial Intelligence: A Modern Approach,” Prentice Hall, New Jersey(2003).
[28] Landwehr, N., Hall, M., and Frank, E., “Logistic Model Trees,” Machine Learning. Vol. 95, No. (1-2), pp. 161-205(2005).
[29] Friedman, J., Hastie, T., and Tibshirani, R., “Additive Logistic Regression: a Statistical View of Boosting,” The Annals of Statistics, Vol. 28, No. 2, pp. 337-407(1998).
[30] Perlich, C., and Provost, F., “Tree Induction vs Logistic Regression,” The Journal of Machine Learning Research, Vol. 4, pp. 211-255(2003).