本研究主要針對影像感測元件封裝製程中之玻璃蓋片瑕疵進行檢測與分類。玻璃蓋片檢測分為兩個部分，分別為非光阻區檢測與光阻區檢測。非光阻區瑕疵有暗點、溶液殘留與刮傷，而光阻區有光阻殘缺與崩邊等瑕疵。
本研究首先針對非光阻區部分以平滑濾波的方式保留影像光源分佈並大幅降低瑕疵信號產生一平滑影像，接著利用原影像與其相減
，去除光源分佈並留下瑕疵影像，再搭配影像二值化、形態學(Morphology)，進行瑕疵檢測(Defect Detection)。此檢測流程經驗證可成功取得影像上之溶液殘留、暗點與刮傷瑕疵，將檢測出之瑕疵尺寸與原影像瑕疵比較後顯示此流程對於低對比度瑕疵如溶液殘留，有良好的效果，而對於高對比度瑕疵如暗點，可明顯的顯示產生瑕疵放大的現象，因此本研究再提出以最小矩形區域(Bounding Box)的方式取出高對比度瑕疵，並使用影像伸張強化提高對比度後，進行影像再分割，成功的取得與原始瑕疵影像尺寸相近的二值化瑕疵影像，達到微觀的玻璃蓋片瑕疵檢測，以實際標準的樣本測試發現可檢測出最小面積2.5μm 2.5μm的暗點，驗證本研究成果符合業界對玻璃蓋片所制定之5μm 5μm瑕疵大小檢測規範。接著對檢測到的瑕疵進行影像特徵擷取，主要提取特徵有平均灰階值、面積、似圓性、灰階均勻度，並利用決策樹J48演算法進行辨識，其整體辨識率達98.34%。最後針對光阻區部分，以適應性模板遮罩(Adaptive Template Mask)的方式搭配Otsu法與交集運算、差集運算，可達到理想的光阻殘缺與崩邊瑕疵檢出效果，且利用決策樹(Decision Tree)J48演算法對其進行辨識
，可達100%之辨識率。
在本研究中，處理單張像素大小為2592 1944之影像，其非光阻區檢測時間為0.98秒，光阻區檢測時間為0.81秒，總處理時間為1.79秒，解決了人力檢測上耗時的問題，此一系統性的瑕疵檢驗方式能全面的應用在影像感測元件封裝製程中玻璃蓋片瑕疵檢測與分類，提高產品之良率。

This study focused on the glass cover chip defect detection and classification in the image sensor packing process. Glass cover chip defect detection can be divided into two parts, namely, the non-photoresist zone detection and photoresist zone detection. Defects in the non-photoresist zone include dark spots, solution residuals and scratches, and the photoresist zone has defects including photoresist deformity and chipping.
Regarding the non-photoresist zone, this study used the smooth filtering approach to retain the image light source distribution and considerably reduce defect signals to generate a smooth image. Then, by subtracting it with the original image, the light source distribution was eliminated, while the defect image was retained for defect detection using binarization and morphology. The results confirmed that the detection process can be successfully applied in getting the image defects of solution residuals, dark spots and scratches. The comparison of the detected defect dimensions and the original image defects suggested that the process has very good effects on low contrast defects, such as solution residuals. Regarding high contrast defects such as dark spots, the amplification of defects can be apparently highlighted. Therefore, this study determined the high contrast defects using the minimum rectangular regions, and used the image zooming to enhance the contrast for image re-segmentation. The binary defect image in dimensions similar to the original defect image sizes was obtained for micro glass cover chip defect detection. The test on real standard samples found that it can detect the dark spot of minimum area at 2.5μm, which met the industrial standards of the glass cover chip defect sized at 5μm. This stud also extracted the image characteristics of the detected defects. The main characteristics of capturing included the average grey scale value, area, circularity, gray-scale uniformity. The decision-making tree J48 algorithm was used for recognition and the overall recognition rate reached 98.34%. Finally, regarding the photoresist zone parts, the adaptive template mask approach, coupled with the Otsu approach, the intersection computing, and the difference set computing, was used to realize the ideal photoresist deformity and chipping defect detection effects. The decision making tree J48 algorithm was used for recognition, and achieved a recognition rate of 100%.
In this experiment, an image of pixel sized 2592*1944 was processed. The non-photoresist zone detection time was 0.98 sec, the photoresist zone detection time was 0.81 sec, and the total processing time was 1.79 sec. Thus, the time consuming problem of manual detection was solved. The systematic defect detection approach could be fully applied in the detection and classification of glass cover chip defects in the image sensing component packaging process to improve product yield.

[1] 陳宗達，CMOS玻璃蓋片自動光學檢測機台之設計及開發，國立交通大學，工業工程與管理學系碩士班，碩士論文，2005年。
[2] Daut D. G. and Zhao D.,“A flaw detection method based on morphological image processing”, IEEE Transactions on Circuit and System for Video Technology, Vol. 3, No. 6, pp. 389-398(1993).
[3] Lin H. D., “Automated defect inspection of light-emitting diode chips using neural network and statistical approaches”, Expert Systems With Applications, Vol. 36, No. 1, pp. 219-226(2009).
[4] Sokolov S. M. and Treskunov A. S.,“Automatic vision system for final test of liquid crystal displays”, IEEE International Conference on Robotics and Automation, Moscow, Vol. 2, pp. 1578-1582 (1992).
[5] Nakashima K.,“Hybrid inspection system for LCD color filter panels”, Tenth International Conference on Instrumentation and Measurement Techonology, Hamamatsu, pp. 689-692(1994).
[6] Liu W. W., Yan Y. H., Li J. , Zhang Y. and Sun H. W.,“Automated on-line fast detection for surface defect of steel strip based on multivariate discriminant function”, Intelligent Information Technology Application, Vol. 2, pp. 493-497(2008).
[7] Chang C. L., Chang H. H. and Hsu C. P., “An intelligent defect inspection technique for color filter”, IEEE International Conference on Mechatronics, pp. 933-936(2005).
[8] Luc V. and Soille P.,“Watersheds in digital spaces: an efficient algorithm based on immersion simulations”, IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 13, No. 6, pp. 583-598(1991).
[9] 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).
[10] Johannsen G. and Bille J., “A threshold selection method using information measures”, Proc. 6th Int. Conference Pattern Recognition Munich, Germany, pp. 140-143(1982).
[11] 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, pp. 273-285(1985).
[12] Tsai W., “Moment-preserving thresholding: A new approach”, Computer Vision, Graphics, and Image Processing, Vol. 29, pp. 377-393(1985).
[13] Ridler T. W. and Calvard S., “Picture thresholding using an iterative selection method”, IEEE Transactions on Systems, Man and Cybernetics, Vol. 8, No. 8, pp. 630 - 632 (1978).
[14] Chang C. Y., Li C. H., Chang Y. C. and Jeng M. D., “Automatic die inspection for post-sawing LED wafers”, Proceedings of the 2009 IEEE International Conference on Systems, San Antonio, pp. 1615-1620( 2009).
[15] Acciani G., Brunetti G. and Fornarelli G., “Application of neural networks in optical inspection and classification of solder joints in surface mount technology”, IEEE Transaction on Industrial Informatics, Vol. 2, NO. 3,pp. 1551-3203(2006).
[16] Frank E., Hall M. A., Holmes G., Kirkby R. B., Pfahringer B. and Witten I.H., “Data mining and knowledge discovery handbook springer”, pp. 1305-1314(2005).
[17] 繆紹綱等，數位影像處理活用Matlab，全華圖書股份有限公司，2010年。
[18] 連國珍等，數位影像處理Matlab，儒林圖書有限公司，2009年。
[19] Carsten S., Markus U. and Christian W.等，機器視覺演算法與應用(繁體中文版)，新亞洲儀器股份有限公司，2011年。
[20] 鍾國亮，影像處理與電腦視覺(第四版)，東華書局，2008年。
[21] Bouaynaya N., “Theoretical foundations of spatially-variant mathematical morphology part II: Gray-level images”, IEEE Trans. on Pattern Analysis and Maxhine Intelligence, Vol. 30, pp. 831-850(2008).
[22] Bratko I. and Muggleton S., “Applications of inductive logic programming”, communications of the ACM, Vol. 38, No. 11, pp. 65-70(1995).
[23] Han J. and Kamber M., “Data Mining: Concepts and techniques”, Morgan Kaufmann, New York(2001).
[24] Russell S. and Norvig P., “Artificaial Intelligence: A modern approach“, Prentice Hall, New Jersey(2003).
[25] Quinlan J. R., “C4.5: Programs for machine learning”, Morgan Kaufmann Publishers, San Mateo, CA(1993).
[26] Quinlan J. R., “Induction of decision tress, machine learning”, pp. 81-106(1986).
[27] Shannon C. E., “A mathematical theory of communication”, Bell System Technical Journal, Vol. 27, pp. 379-423(1948).