本研究旨在以影像處理方法，對晶圓級顯微影像進行樣板比對。由於晶圓級顯微影像中可能出現製程瑕疵，而使樣本影像與樣板影像有所差異。現有影像樣板比對方法中，關鍵點特徵對應方法以多點局部特徵，將影像位置予以對應，在影像中存有製程瑕疵之前提下，該方法最為適用。
關鍵點特徵比對方法以兩張影像中的關鍵點進行對應，需要依賴影像中的特定局部結構做為關鍵點，當同一影像中之多數關鍵點特徵相似，將會使對應具有模糊性。針對該缺點，本研究提出旋轉不變特徵匹配法，在不依賴影像中特定結構之前提下，進行影像之特徵對應。
旋轉不變特徵匹配法整合了影像加速技巧與資料結構演算法，縮短建立與對應影像特徵的計算時間。本研究以T公司提供之晶圓級顯微影像進行試驗，將關鍵點特徵方法之對應失敗影像，以旋轉不變特徵方法進行測試，能夠全數對應成功，對應解析度916×1280之影像，平均一張需時約27.68秒，誤差平均值為2.05個畫素寬度。

The research is focused on the topic of template matching with microscopic image at wafer level by image processing methods. Due to the occurrence of defect in the wafer, the template and the sample wafer-level microscopic image might be different. Among existing template matching methods, key point feature matching methods match image position by local feature of multiple locations.
Key point feature matching method define specific local structure as keypoint. Matching ambiguity occurs when most of the keypoint feature are similar. Against this disadvantage, we propose rotation-invariant feature matching method, which matches images independent of any specific local structure.
Combined with speed-up image processing method and data structure analysis method, we reduce the computation time of building and matching image features. Our research tested rotation–invariant feature matching method with wafer-level microscopic image provided by TSMC which was tested and failed by using keypoint matching method. Our method can successfully match all images among them. The average computation time of image matching by our method is 28.24 second, with only 2.05 pixel average geometric error.

摘要 i
Abstract ii
致謝 iii
圖目錄 vii
表目錄 xii
第1章 簡介 1
1.1 問題背景與描述 3
1.2 文獻回顧 4
1.2.1 影像區域對位 4
1.2.2 關鍵點特徵匹配 5
1.2.3 旋轉不變特徵對應 8
1.3 研究策略與論文架構 9
第2章 關鍵點特徵匹配法 11
2.1 關鍵點搜尋 11
2.1.1 影像座標空間之關鍵點搜尋 11
2.1.2 尺度空間之關鍵點搜尋 13
2.1.3 尺度不變特徵轉換 – 高斯差 16
2.1.4 加速強健特徵 – 積分影像 19
2.1.5 方法比較 22
2.2 特徵描述 23
2.2.1 尺度不變特徵轉換 23
2.2.2 加速強健特徵 25
2.2.3 方法比較 27
2.3 小結 27
第3章 旋轉不變特徵匹配法 28
3.1 環形投影特徵 30
3.2 樣板影像之關鍵點搜尋 36
3.2.1 KD樹之建立 38
3.2.2 KD樹之最鄰近搜索 40
3.3 特徵對應 45
第4章 排除錯誤對應 46
4.1 隨機抽樣共識法 47
4.2 全域共識法 49
第5章 實驗與結果討論 52
5.1 參數分析 56
5.1.1 參數分析-特徵向量維度 57
5.1.2 參數分析-最外圈半徑 59
5.1.3 參數分析-關鍵點數量 61
5.1.4 參數分析-關鍵點取樣間隔 64
5.2 與現有方法的比較 66
5.2.1 旋轉不變性-大角度旋轉 67
5.2.2 旋轉不變性-小角度旋轉 69
5.2.3 抗雜訊強健度 71
5.2.4 效能比較 74
第6章 結論 76
6.1 本研究之貢獻 76
6.2 未來展望 77
參考文獻 78

[ 1 ]M. Brown and D.G. Lowe, “Automatic panoramic image stitching using invariant features,” International Journal of Computer Vision, vol. 74, pp. 59-73, 2007.
[ 2 ]S.C. Park, M.K. Park and M.G. Kang, “Super-resolution image reconstruction: a technical overview,” IEEE Signal Processing Magazine, vol. 20, pp.21-36, 2003.
[ 3 ]A. Jiang, D.N. Kennedy, J.R. Baker, R.M. Weisskoff, R.B.H. Tootell, R.P. Woods, R.R. Benson, K.K. Kwong, T.J. Brady, B.R. Rosen and J.W. Belliveau, “Motion detection and correction in functional MR imaging,” Human Brain Mapping, vol. 3, pp. 224-235, 1995.
[ 4 ]B.D. Lucas and T. Kanade, “An Iterative Image Registration Technique with an Application to Stereo Vision,” Proceedings of Imaging Understanding Workshop, vol. 3, pp. 121-130, 1981.
[ 5 ]S. Baker and I. Matthews, “Lucas-kanade 20 years on: A unifying framework,” International Journal of Computer Vision, vol. 56, pp.221-255, 2004.
[ 6 ]A. Bartoli, “Groupwise Geometric and Photometric Direct Image Registration,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 30, pp.2098-2108, 2008.
[ 7 ]H. Moravec, “Obstacle Avoidance and Navigation in the Real World by a Seeing Robot Rover,” Tech Report CMU-RI-TR-3 Carnegie-Mellon University, Robotics Institute, 1980.
[ 8 ]O.A. Zuniga and R.M. Haralick, “Gradient Threshold Selection Using the Facet Model,” Proceeding of Pattern Recognition and Image Processing, vol. 21, pp. 30-37, 1983.
[ 9 ]C. Harris and M. Stephens, “A combined corner and edge detector,” Proceedings of the 4th Alvey Vision Conference, vol. 1, pp. 147–151, 1988.
[ 10 ]X. Dai and S. Khorram, “The effects of image misregistration on the accuracy of remotely sensed change detection,” IEEE Transaction on Geoscience Remote Sensing, vol. 36, pp. 1566–1577, 1998.
[ 11 ]T. Lindeberg, “Feature detection with automatic scale selection,” International Journal of Computer Vision, vol. 30, pp.79-116, 1998.
[ 12 ]D.G. Lowe, “Distinctive Image Features from Scale-Invariant Keypoints,” International Journal of Computer Vision, vol. 60, pp. 91-101, 2004.
[ 13 ]H. Bay, A. Ess, T. Tuytelaars and L. Van Gool, “Speeded-Up Robust Features (SURF),” Computer Vision and Image Understanding, vol. 110, pp.346-359, .
[ 14 ]M.K. Hu, “Visual pattern recognition by moment invariants,” IRE Transactions on Information Theory, vol. 1, pp.179-187, 1962.
[ 15 ]C.H. Teh and R.T. Chin, “On image analysis by the methods of moments,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 10, pp.496-513, 1988.
[ 16 ]F. Ullah and S. Kaneko, “Using orientation codes for rotation-invariant template matching,” Pattern Recognition, vol. 37, pp.201-209, 2004.
[ 17 ]D. Marimon and T. Ebrahimi, “Efficient rotation-discriminative template matching,” 12th Iberoamerican Congress on Pattern Recognition, vol. 4756, pp.221-230, 2007.
[ 18 ]H.Y. Kim and S.A. Araújo, “Grayscale template-matching invariant to rotation, scale, translation, brightness and contrast,” IEEE Pacific-Rim Symposium on Image and Video Technology, vol. 4872, pp.100-113, 2007.
[ 19 ]J.L. Bently, “Multidimensional Binary Search Trees Used for Associative Searching,” Communication of the ACM, vol. 18, pp.509-517, 1975.
[ 20 ]M.A. Fischler and R.C. Bolles, “Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography,” Comunications of the ACM, vol. 24, pp. 381-395, 1981.
[ 21 ]
http://www.vlfeat.org/overview/sift.html, 2012, June.

[ 22 ]
http://www.mathworks.com/matlabcentral/fileexchange/28300-opensurf-including-image-warp, 2013, February.

[ 23 ]
https://sites.google.com/site/andreatagliasacchi/software/matlabkd-treelibrary, 2012, January.