- 學位論文
多階混合式光學字元辨識技術與分析
Multi-Stage Hybrid Optical Character Recognition for IC marks
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
在光學字元辨識實際工業應用中，外在的光源、字元尺寸、字型和位移旋轉都是影響辨識結果的因素，因此本研究探討相關係數法﹙correlation coefficient﹚、樣板比對法﹙template matching﹚、環形投影法﹙ring projection﹚、細線化法﹙thinning﹚和本研究所提出的四分法（quartering）五種辨識方法，利用計算字元的相似度係數，以進行各種影響因子的分析，並比較五種方法受到不同因素影響後的辨識能力。 本研究主要針對IC字元辨識，藉由探討目前多種光學字元辨識技術，分析各種辨識方法受到環境因素影響的程度，並且根據各種辨識方法的特性，進而提出一個結合現有辨識技術，建立階段式的辨識步驟，以解決只使用單一種字元辨識方法所產生的字元誤判情況。利用字元間的相似度係數信賴區間（confidence interval）箱形圖（box plots）的方法，將字元箱形圖有重疊者之字元分為同一群組，針對每一個階段中之每個群組內的字元再選用相對適合的辨識方法以建立子分群，使得字元間有明顯的差異性，重複此分群步驟直至每一字元成為單獨一個字群。進行辨識時，是依據已經建立好之分群模式，在每一階段中之每個群組內使用選定的辨識方法進行辨識。使用本研究所提之多階混合式辨識策略，可以提升字元辨識成功率，避免誤判字元的發生。
並列摘要
In this research we aim at the optical character recognition (OCR) techniques for IC marks in electronic industry. We analyze four existing OCR techniques including correlation coefficient method, template matching method, ring projection method, and thinning method, and a new quartering method proposed in this research. These five OCR techniques evaluated are computationally simple and easily implemented. The effects of lighting, font, scale, translation and rotation on the similarity measures are evaluated for each OCR method. The characters of IC marks include 26 upper-case alphabets and numerical 10 digits. Since none of the five OCR methods aforementioned can reliably recognition all 36 characters, this research employs a multi-stage, hybrid-method strategy to maximize the recognition rate. In the training phase, the correlation coefficient method is initially used to group similar characters into a subgroup. Characters in the subgroup are further partitioned into smaller group using one of the remaining methods. The partition procedure is recursively repeated until there is only one character in each subgroup. The selected OCR method for each character group in each stage is then used for character classification in the recognition phase. Experiments on various IC marks have shown that the proposed algorithm can achieve 100% recognition rate.
參考文獻
Borgefors, G., 1988, "Hierarchical chamfer matching: A parametric edge matching algorithm", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 10, No. 6, pp. 849-865.
Cash, G. L., and Hatamian, M., 1987, "Optical character recognition by rhe method of moment", CVGIP, Vol. 39, pp. 291-310.
Chaudhuri, B. B. and Pal, U., 1998, "A complete printed Bangla OCR system", Pattern Recognition, Vol. 31, No. 5, pp. 531-549.
Cordella, L. P., Stefano, C. D., and Vento, M., 1995, "A neural network classifier for OCR using structural descriptions", Machine Vision and Applications, Vol. 8, No. 5, pp. 336-342.
Hong, C., Loudon, G., Wu, Y., and Zitserman, R., 1998, "Segmentation and recognition of continuous handwriting Chinese text", International Journal of Pattern Recognition and Artificial Intelligence, Vol. 12, No. 2, pp. 223-232.