臺灣博碩士論文加值系統

現今電子產品的普及，使得各種精密電子零件的需求量大增，相對地，產品瑕疵的檢測和尺寸精密度的要求也越來越高，因此人工檢測方式已漸漸不敷要求。因此光學自動檢測技術(AOI)也日漸受到重視，雖然此技術已廣泛應用在工業產品的生產檢測上，但就精密零件檢測方面的應用來說，目前大部分的技術仍須以人工方式尋找物件影像的位置，並加以圈選辨識，或是需限制同一時間檢測的待測物其形狀種類和尺寸的範圍，對於自動化檢測技術來說，尚有許多的不足之處，但同時也還有很大的改進空間。
目前針對幾何辨識的工作主要還是使用Hough Transform來尋找幾何參數，因為Hough Transform是針對所有可能的幾何形狀參數進行搜尋的工作，因此需要事前得知參數範圍或只能進行小尺寸和小區域的參數搜尋工作，以避免計算量過大的問題，而現有的技術中，針對一整張640*480像素大小的影像進行Hough Transform時，需要事先限制參數變化的範圍，否則將會產生計算量過大的問題，也因此限制了Hough Transform的應用範圍；另外，也有先利用影像尺寸壓縮的技術，對縮小後的影像得到初步的參數後再疊代計算得到原尺寸影像的幾何參數，但此技術的缺陷在於尺寸壓縮的技術會造成細微資訊的遺漏，使得辨識結果大打折扣。
本論文提出利用影像分割Labeling加上轉角偵測的方法，改善原有利用影像尺寸壓縮來搜尋幾何參數的辨識技術。使用影像分割Labeling的方式，可以快速定位出影像中待測物的位置，同時利用轉角偵測將待測物的邊緣輪廓切割成基本圖形，以便後續針對個別的基本圖形進行尺寸檢測，如此能更快速且精確地完成檢測工作。
本文所提出的幾何辨識流程為：先對由CCD擷取的整張影像進行邊緣點偵測後，再進行Labeling分組的動作，將影像中的待測物的邊緣點,依照連結的情形,分成不同的物件群組，接著再利用偵測轉角的方式將群組中的物件輪廓切割成基本的幾何形狀，並對分割出的基本形狀計算其面積慣性矩(moment of inertia)，分辨出最有可能呈現直線和圓分佈的邊緣點，最後分別對不同幾何形狀的邊緣點採用相對應的辨識方法，如此一來可減少計算次數，同時減少雜訊的干擾，提高辨識精準度。
在一般的機械加工零件的檢測中，最常檢測的元件都是由基本的幾何形狀所組成，因此，使用本論文所提的的影像分割辨識方法，將可以改善現有技術無法同時檢測多種幾何形狀的幾何參數的限制，同時對於事先無法得知物件位置的影像，也可以快速的將物件定位並找出幾何參數，可大幅提高檢測速度，同時維持一定的檢測品質。

Due to the prevailing of the miniature electronic products, demands for precision parts for these products have increased dramatically over the last decade. Due to their small dimensions and total, instead of partial, product inspection, have made manual inspection method unfeasible. Meanwhile, requirements on the automatic optical inspection (AOI) technology to detect product defect and ensure the dimensional accuracy become more stringent.
Although, AOI technology has been applied to the inspection of industrial product for decades. The inspection task, for example, firstly to locate the position and let the vision software gauge the radii of holes in a mechanical component. The first step is mostly carried out manually. This limits the performance of the AOI equipments.
This paper proposes a geometric parameter extraction method using the image segmentation technique to improve an existing method using image size reduction in searching geometrical parameters. Image segmentation can help us to quickly locate the positions of all objects in an image, and corner detection segment each object into basic geometrical elements like lines, arcs and circles. Then we can choose the corresponding geometrical parameters extraction’s method to find the geometrical parameters of the pattern. There will be more precise and fast in searching geometrical parameters.
The process of geometrical inspection is first to detect edge, then to use edge linking and labeling to classify existing objects. Following, utilizing the characteristics of SUSAN edge response to locate the corners, and the object of the first classification will reclassify again. Then the geometric shape in original image can be segmented in basis shape, and we can calculate the inertial value for every segment to recognize the circles and lines. Finally, using correspond method to each segmented group. Hence, it can prevent time wasting and reduce disturbance to enhance the precision of the geometrical parameter inspection.

第 1 章 緒論.......................................................1
1.1 研究動機...................................................1
1.2 文獻探討...................................................2
1.3 論文大綱...................................................5
第 2 章 理論探討...................................................6
2.1 影像相鄰像素標記(Labeling).................................6
2.2 轉角偵測的研究.............................................8
2.3 SUSAN的原理介紹...........................................10
2.2.1 計算USAN面積值和SUSAN Edge Response.......................11
2.2.2 邊緣點偵測................................................14
2.2.3 SUSAN轉角偵測.............................................16
2.4 曲率尺度空間Curvature Scale Space原理介紹.................19
2.3.1 曲率尺度空間(Curvature Scale Space ,CSS)..................19
2.3.2 CSS轉角偵測...............................................20
2.5 本文尋找轉角的方式........................................22
2.6 幾何形狀判別..............................................25
2.7 計算幾何參數..............................................26
2.7.1 Hough Transform...........................................26
2.7.2 Curve Fitting.............................................29
2.8 幾何形狀辨識流程..........................................33
2.8.1 使用SUSAN轉角偵測.........................................33
2.8.2 使用CSS轉角偵測...........................................34
第 3 章 實驗與討論................................................36
3.1 Labeling標籤分組結果......................................37
3.2 USAN面積結果..............................................39
3.3 SUSAN邊緣點偵測結果.......................................40
3.4 CSS對邊緣輪廓的影響.......................................42
3.5 SUSAN和CSS轉角偵測結果....................................44
3.5.1 SUSAN轉角偵測結果.........................................44
3.5.2 CSS轉角偵測結果...........................................45
3.6 SUSAN和CSS等轉角理論應用於實際影像的結果..................47
3.7 利用轉角特性分出直線......................................48
3.7.1 SUSAN理論尋找轉角.........................................48
3.7.2 CSS理論尋找轉角...........................................50
3.8 進行幾何辨識.............................................52
3.8.1 計算Inertia結果...........................................52
3.8.2 圓形幾何參數辨識..........................................53
3.8.3 直線幾何參數辨識..........................................54
3.8.4 圓弧幾何辨識..............................................55
3.9 使用CSS轉角偵測辨識流程於實際影像.........................57
3.9.1 實際影像一................................................58
3.9.2 實際影像二................................................64
3.10 結果與討論................................................71
第 4 章 結論......................................................73
附錄..............................................................77
1.應用SUSAN理論的幾何辨識方法於一張640*480像素的測試圖結果........77
2.使用CSS轉角理論的幾何辨識方法於更多測試圖的結果.................81

[1]. L.G. Roberts, “Machine Perception of Three-Dimensional Solids,” in J. T. Tippet, Editor, Optical and Electro-optical Information Processing, pp. 159-197, MIT Press, 1965
[2]. J.M.S. Prewitt, “Object Enhancement and Extraction,” in B.S. Lipkin and Rosenfeld, Editors, Picture Processing and Psychopictorics, Academic Press, pp.75-149, 1970
[3]. J. Canny A “Computational Approach to Edge Detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 8, No. 6, November 1986
[4]. I. Sobel, “An Isotropic Image Gradient Operator,” in H. Freeman, Editor, Machine Vision for Three-Dimensional Scenes, Academic Press, pp. 376-379, 1990
[5]. S.M. Smith and M. Brady. “SUSAN - A New Approach to Low Level Image Processing,” International Journal of Computer Vision, Vol. 23, No. 1,, pp. 45-78, 1997.
[6]. H. P. Moravec. “Towards Automatic Visual Obstacle Avoidance,” Proceedings. 5th International Joint Conference on Artificial Intelligence, pp. 584, 1977.
[7]. H. P. Moravec, “Visual Mapping by a Robot Rover,” International Joint Conference on Artificial Intelligence, pp. 598-600, 1979.
[8]. C. Harris and M. Stephens, “A Combined Corner and Edge Detector,” Proceedings of Alvey Vision Conference, Manchester, pp. 147-151, 1988.
[9]. M. Trajkovic and M. Hedley. “Fast Corner Detection,” Image and Vision Computing, Vol. 16, No.2 , pp. 75-87, 1998. [10]. F. Mokhtarian, and R. Suomela, “Robust Image Corner Detection through Curvature Scale Space,” IEEE Transactions. Pattern Analysis and Machine Intelligence, vol. 20, number 12, pp. 1376-1381, 1998.
[11]. F. Mokhtarian, and F. Mohanna, “Enhancing the curvature scale space corner detector,” Proc. Scandinavian Conference on Image Analysis, pp. 145-152, Bergen, Norway, 2001.
[12]. He, X.C. and Yung, N.H.C., “Curvature scale space corner detector with adaptive threshold and dynamic region of support,” Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on Volume 2, 23-26 Aug, 2004
[13]. Clark F. Olson, “Locating Geometric Primitives by Pruning the Parameter Space,” Pattern Recognition, Vol. 34, pp. 1247-1256, 2001
[14]. H. Li, M. A. Lavin, R. J. Le Master, “Fast Hough Transform: A Hierarchical Approach,” Computer, Vision Graphics, Image Processing, Vol. 36, pp. 139-161,1986
[15]. S. Y. Yuen, K. H. Lo, “A Coarse-to-Fine Approach for Circle Detection,” Proc. Int. Symposium on Speech, Image Processing & Neural Networks, Hong Kong, pp. 523-526, April 1994
[16]. L. Xu, E. Oja, “Randomized Hough Transform (RHT): Basic Mechanism, Algorithms, and Computational Complexities,” CVGIP: Image Understanding, Vol. 57, No. 2, pp. 131-154, 1993
[17]. H. Kalviainen, “Randomized Hough Transform: New Extensions,” Lappeenranta University of Technology, 1994
[18]. Corneliu Rusu, “Classical geometrical approach to circle fitting – review and new developments,” Journal of Electronic Imaging, Vol. 12, No. 1, pp. 179-193, 2003
[19]. 涂軍橋,”精密零件幾何形狀辨識,” 元智大學,機械工程學系,碩士,2003