本論文提出以重心導引為基礎之改良型密度分群演算法（GDBSCAN：Gravity-Orientated Density-Based Clustering of Application with Noise），此方法將演算法之檢視區域劃分為四個象限，以資料分布密度求出重心象限，並以此重心象限導引演算法成長之方向，進行以密度為基礎之資料點分群過程。本研究將所提出之方法搭配影像處理技術應用於抗反射玻璃（AR Glass：Anti-Reflection Glass）之瑕疵檢測技術，其檢測項目包括雜點、亮點以及刮痕瑕疵檢測。
擷取到抗反射玻璃之影像後，利用拉普拉斯運算子將影像強化，突顯影像中瑕疵資訊部分，接著採用平滑法能抵抗雜訊之能力將影像中不必要之資訊消除，得到改善後之影像後，使用適應性二值化將影像進行分割，將瑕疵資訊從背景中分離出來，最後利用本論文所提出之GDBSCAN將瑕疵資料點進行分群之動作，由實驗結果可知，本方法可將300*200 pixels之AR玻璃局部影像成功地分群並標記出瑕疵群落之座落之處及大小範圍，且每張300*200大小的AR玻璃檢測流程平均花費21毫秒。當影像資料點數大於7000點時，由演算法運算時間表可知本方法比 FDBSCAN 分群演算法平均有17%的節省時間率，對於RDBSCAN 平均有10%節省時間比例。

This thesis proposes Gravity- Orientated Density-Based Clustering Applications with Noise（GDBSCAN）. GDBSCAN separates the query region into four quadrants and calculates each quadrant density individually. The quadrant with the highest density is called Gravity-quadrant. Gravity-quadrant will orientate the expending direction of the density-based clustering algorithm. The proposed method and image process technology are adopted for AR glass defects detection. The detected items include impurities, bright dots and scratch.
During the inspection process, a Laplacian operator is employed to enhance the image, firstly. Then, a smoothing process is also taken to reduce the noise before binary process. Subsequently, an adaptive threshold technique is applied to extract the defect images from background of AR glass image. Finally, the AR glass defects is clustered by the proposed GDBSCAN. According to the experimental results, the GDBSCAN can successfully discover the defects of a AR glass image of 300*200 pixels. And it totally takes 21 ms for defects detecting process of each AR glass. The GDBSCAN can save 17% execution time as compared to FDBSCAN and 10% compared to RDBSCAN for an image with higher than 7000 defect points.

[1] 電子工程專輯， http://www.eettaiwan.com/ART_8800568669_480702_NT_3db86bee.HTM ，2007。
[2] 岳華展光電股份有限公司， http://www.rocoes.com.tw/， 2008/03。
[3] 陳燕儀、呂世源，「光學膜-光的魔術師」科學發展，414 期，2007。
[4] 坤輝科技實業股份有限公司 ，http://www.ub-chem.com/，2009。
[5] 趙新民，“應用非線性擴散於非同質性紋路之ITO導電玻璃表面檢測”，元智大學工業工程與管理學系碩士論文，2003。
[6] 曾彥馨，「應用機器視覺於 TFT面板之表面瑕疵檢測與分類」，碩士論文，元智大學工業工程與管理學系碩士論文，2003。
[7] 洪崇祐，「應用一維傅立葉分析於TFT-LCD液晶顯示面板之瑕疵檢測」，元智大學工業工程與管理學系碩士論文，2004。
[8] 游舜豪，「應用高解析度線掃描CCD於ITO導電玻璃表面瑕疵檢測之研 究」， 碩士論文，台灣科技大學自動化及控制研究所碩士論文，2006。
[9] Tsai, D. M. and T. Y. Huang, 2003, “Automated Surface Inspection for Statistical
Textures,＂ Image and Vision Computing, Vol. 21, pp. 307-323.
[10] Der-Baai Perng,Cheng-Chuan Chou ,Wei-Yu Chen,＂A Novel Vision for CRT Panel Auto-inspection＂International Conference on Mechatronixs, July10-12, IEEE, 2005.
[11] 謝承恩，「以累積差異及多重解析度背景相減偵測 LCD Mura 瑕疵」， 國立中央大學資訊工程研究所碩士論文，2007。
[12] 林宏達，「透明玻璃之瑕疵偵測與瑕疵之正反面位置判斷」，朝陽科技大學工業工程與管理所碩士論文，2007。
[13] 林祺銘，「玻璃基板邊緣檢測系統之研發」，國立高雄應用科技大學機械與精密工程研究所碩士論文，2008。

[14] 李坤憙，「信用卡與玻璃基板之刮痕檢測」，南台科技大學電子工程研究所碩士論文，2006。
[15] 吳宗諭，「一種以密度為基礎之改良型快速分群演算法及即時偏光板瑕疵區域檢測之應用」，台灣科技大學自動化及控制研究所碩士論文，2006。
[16] J. MacQueen, “Some Methods for Classification and Analysis of Multivariate Observations, ” In Proc. 5th Berkeley Symp. Math. Stat. and Prob., Vol. 1, pp. 281-297, 1967.
[17] Guha, Sudipto, Rajeev Rastogi, and Kyuseok Shim, “ROCK: A Robust Clustering Algorithm for Categorical Attributes,” IEEE, Conference on Data Eng., 1999
[18] M. Ester, H.P. Kriegel, J. Sander, X. Xu, “A density-based algorithm for discovering clusters in large spatial databases with noise, ” Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, pp. 226-231, 1996.
[19] W. Wang, Yang, R. Muntz, STING: A Statistical Information grid Approach to Spatial Data Mining, VLDB’97
[20] 林育臣，“群聚技術之研究”，朝陽科技大學資訊管理研究所碩士論文，2002。
[21] Bing Liu, “A Fast Density-Based Clustering Algorithm for Large Databases, ” Machine Learning and Cybernetics, pp. 996-1000, August, 2006.
[22] M.Ester, Kriegel H.P., Sander J., Wimmer M. and Xu X., Aug.1998,“Incremental Clustering for Mining in a Data Warehousing Environment,” InProc. 24th Int. Conf. Very Large Databases (VLDB'98), pp. 24 - 27, New York.
[23] M. Ankerst, Breunig M., Kriegel H.P., Sander J. OPTICS: Ordering points to identify the clustering structure. ACM SIGMOD Int. Conf. on Management of Data, Philadelphia, PA, pp. 49-60. , 1999.
[24] B.Borah, , Bhattacharyya, D.K., “An Improved Sampling-Based DBSCAN for Large Spatial Databases, ” Intelligent Sensing and Information Processing, pp. 92-96, 2004.
[25] Daoying Ma and Aidong Zhang, “An Adaptive Density-Based Clustering Algorithm for Spatial Database with Noise, Data Mining, 2004. ICDM '04. Fourth IEEE International Conference. pp.467 – 470,2004.
[26] P. Viswanath, Pinkesh, R., “l-DBSCAN : A Fast Hybrid Density Based Clustering Method, ” Pattern Recognition, 2006. ICPR 2006. 18th International Conference on Volume 1, pp. 912-915, 2006.
[27] B. Borah' and D. K. Bhaftacharyya, “ A Clustering Technique using Density Difference” Signal Processing, Communications and Networking, pp. 585-588, 2007.
[28] Ming-Jong Tsai, Chung-Yu Wu, Chao-Hsiang Feng, F. N. Tien,” A Rapid Density-Based Clustering Algorithm for Large Databases” CVGIP2008,I-lan, Aug. , 2008.
[29] Ming-Jong Tsai, Fen-Ning Tien and Chung-Yu Wu” Automatic Optical Inspection of Anti-reflection Glass by using A Rapid Density-Based Clustering Algorithm” Automation 2009, Tainan, June, 2008.
[30] 台灣仿真科技股份有限公司，http://www.jubilee.com.tw/，2009。
[31] 鐘國亮，”影像處理與電腦視覺”，東華書局，2006。