目前偏光膜瑕疵是利用人工檢測的方式，不僅費時而且容易因疲勞而產生檢測失誤，因此第三者客觀公正的專業檢測服務將可做為重要依據，所以本論文針對偏光膜外觀瑕疵應用影像處理來開發一套自動瑕疵檢測系統，而檢測的偏光膜瑕疵可分為五種：包括雲狀色差、條狀色差、點狀色差、刮痕與貼合不良。在影像處理過程中，先利用中值濾波器減少脈衝雜訊，由於瑕疵影像灰階差異不大，無法以一固定門檻值完整分割所有瑕疵類別，故使用統計式門檻值決定法配合灰階值大小，以選擇一個或兩個最佳門檻值來分割出瑕疵區域，再配合形態學中的閉合運算使瑕疵輪廓更平滑完整，並選擇瑕疵面積、平均灰階值和緊緻性做為瑕疵特徵。我們搜集80個瑕疵樣本，並規劃以不同數量之訓練樣本與固定的測試樣本進行實驗，利用20筆、30筆和40筆訓練樣本做為倒傳遞類神經網路的資料庫，對固定的40筆測試樣本作辨識，結果顯示在訓練樣本為30筆以上時，其辨識率皆可達到100%，驗證了倒傳遞類神經網路在訓練樣本足夠的資料庫下，可獲得相當準確的辨識率，成功被應用於偏光膜瑕疵自動檢測系統。

Currently, polarizing film defects are inspected by the human, which is tedious and easily leads to wrong judgments due to tiredness of eyes. Thus the objective inspection becomes an important and necessary basis. This thesis applies image processing and appropriate classifiers to develop an image inspection system for polarizing film appearance defects. The polarizing film defects include cloudy chromatism, stripy chromatism, spotted chromatism, scratch and pasting wrong, which are commonly occurred during processing. In image processing, the median filter is used to reduce the impulse noise of images. Since the gray values of all defects are close to each other, we cannot segment entire defect-areas of five defects by a fix threshold value. Therefore, the statistical threshold value decision method is used to choose one or two optimal threshold value with the difference of gray values in image segmentation, and then we use the closing operator in morphology to smooth the contour of defects. Moreover, the area, average gray value and compactness are selected as defect features. We divide eighty defect samples into three groups, which are the different amount of training samples and the fixed amount of testing samples, for experiment. The amounts of training samples are twenty, thirty and forty, and the amount of testing samples is forty. Finally, we have a database of the back-propagation neural network (BPNN) trained by the training samples and recognize the testing samples with the database. The result shows the recognition rate is 100% when the amount of training samples is greater than or equal to thirty; proving that the BPNN can achieve a high recognition rate with enough training samples in the database, and it can be successfully applied to the inspection of polarizing film defects.

[1]C. Hecht and G. Dishon, “Automatic optical inspection (AOI),” Proceedings of the 1990 40th Electronic Components and Technology Conference, vol. 1, pp. 659-661, 1990.
[2]H. F. Ng, “Automatic thresholding for defect detection,” Pattern Recognition Letters, vol. 27, no. 14-15, pp. 1644-1649, 2006.
[3]J. N. Kapur, P. K. Sahoo and A. K. C. Wong, “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.
[4]C. J. Huang, C. F. Wu and C. C. Wang, “Image processing techniques for wafer defect cluster identification,” IEEE Design & Test of Computers, vol. 19, no. 2, pp. 44-48, 2002.
[5]S. J. Ko and Y. H. Lee, “Center weighted median filters and their applications to image enhancement,” IEEE Trans. on Circuits and Systems, vol. 38, no. 9, pp. 984-993, 1991.
[6]徐國政，皮革瑕疵類別自動辨識與補償，逢甲大學工業工程學系碩士論文，2002。
[7]B. C. Jiang, S. L. Tasi and C. C. Wang, “Machine vision-based gray relational theory applied to IC marking inspection,” IEEE Trans. on Semicond. Manuf., vol. 15, no. 4, pp. 531-539, 2002.
[8]C. L. Lee and S. Y. Chen, “Classification of leaf images,” International Journal of Imaging Systems and Technology, vol. 16, no. 1, pp. 15-23, 2006.
[9]M. Partio, B. Cramariuc, M. Gabbouj and A. Visa, “Rock texture retrieval using gray level co-occurrence matrix,” In 5th Nordic Signal Processing Symposium, 2002.
[10]C. T. Su, T. Yang and C. M. Ke, “A neural-network approach for semiconductor wafer post-sawing inspection,” IEEE Trans. on Semicond. Manuf., vol. 15, no 2, pp. 260-266, 2002.
[11]黃偉碩，利用貝氏分類與因子分析法於半導體製程錯誤偵測與診斷，中華大學科技管理學所碩士論文，2004。
[12]黃朝義，偏光膜介紹，光電產業與技術情報，20期，第40-47頁，1999。
[13]鐘國亮，影像處理與電腦視覺第三版，東華書局，2006。
[14]蔡英男，應用影像處理與類神經網路於偏光膜瑕疵檢測，國立台灣科技大學高分子所碩士論文，2003。
[15]N. Otsu, “A threshold selection method from gray-level histograms,” IEEE Trans. on System, Man and Cybernetics, vol. SMC-9, no. 1, pp. 62-66, 1979.
[16]R. Jain, R. Kasturi and B. G. Schunck, Machine Vision, McGraw-Hill, 1995.
[17]P. B. Chou, A. R. Rao, M. C. Sturzenbecker, F. Y. Wu and V. H. Brecher, “Automatic defect classification for semiconductor manufacturing,” Machine Vision and Applications, vol. 9, no. 4, pp. 201-214, 1997.
[18]葉怡成，類神經網路模式應用與實作，儒林圖書公司，1995。
[19]I. H. Witten and E. Frank, Data Mining: Practical Machine Learning Tools and Techniques (second ed.), Morgan Kaufmann, 2005.
[20]R. O. Duda, P. E. Hart and D. G. Stork, Pattern Classification and Scene Analysis (second ed.), John Wiley & Sons, 2005.
[21]K. Sakurai, A. Onoyama, H. Ishii, K. Oka and K. Yamanishi, “Capture rate enhance method of 0.1-μm level defects by pattern-matching inspectors,” IEEE Trans. on Semicond. Manuf., vol. 13, no. 4, pp. 434-441, 2000.
[22]W. M. Morrow, R. B. Paranjape, R. M. Rangayyan and J. E. L. Desautels, “Region-based contrast enhancement of mammograms,” IEEE Trans. on Medical Imaging, vol. 11, no. 3, pp. 392-406, 1992.
[23]J. W. Lee, B. R. Lim, R. H. Park, J. S. Kim and W. Ahn, “Two-stage false contour detection using directional contrast and its application to adaptive false contour reduction,” IEEE Trans. on Consumer Electronics, vol. 52, no. 1, pp. 179-188, 2006.
[24]黃良華，應用理想解類似度順序偏好法於偏光膜之瑕疵檢測，國立台灣科技大學高分子所碩士論文，2004。
[25]繆紹綱編譯，數位影像處理第二版，普林斯頓國際有限公司，2003。
[26]連國珍，數位影像處理，儒林圖書有限公司，2004。
[27]王進德，類神經網路與模糊控制理論入門與應用，全華圖書公司，2007。
[28]張云濤和龔玲，資料探勘原理與技術，五南圖書公司，2007。
[29]P. Domingos and M. Pazzani, “On the optimality of the simple bayesian classifier under zero-one loss,” Machine Learning, vol. 29, no. 2-3, pp. 103-130, 1997.
[30]M. Skurichina, S. Raudys and R. P. W. Duin, “K-Nearest Neighbors Directed Noise Injection in Multilayer Perceptron Training,” IEEE Trans. on Neural Networks, vol. 11, no. 2, pp. 504-511, 2000.