傳真手寫數字自動辨識系統

Automatic Hand Written Numeral Characters Recognition System for FAX Machine

陳慶彥(Ching-Yean Chen)

指導教授 ： 林啟芳

元智大學/工程學院/電機與資訊工程研究所/碩士(1997年)

摘要

本論文提出方法發展一套傳真手寫數字自動化辨識系統。所提方法以即時方式來辨識傳真文件上的手寫數字，除了可以減少所需的記憶空間來存放整張傳真影像之外，也可節省接收傳真資料時的等待時間。當傳真文件傳送完畢後，本系統即可同時辨識出文件上的手寫數字。在論文中所使用的辨識架構除了主要是自我組織特徵映射類神經網路(SOM)，另外分別設計了不具鄰域學習特性的決策型SOM類神經網路及具鄰域學習特性的決策型SOM類神經網路兩種不同的架構來提昇系統效能。
此外，由於系統在處理時間方面要達到即時的要求，我們採用遺傳演算法(Genetic Algorithm)作最佳化運算，使得系統能夠在不降低整體辨識率的前提下將特徵向量大幅刪減，僅保留有用的特徵組合，提昇了系統辨識效能和執行速度。最後由實驗結果驗證了上述所提方法的可行性與優越性。

關鍵字

手寫數字辨識 ； 類神經網路 ； SOM ； 遺傳演算法則

並列摘要

Using a fax machine to send an order form containing a sequence of handwritten order numbers is a convenient way for people to order or purchase merchandise in recent rears. However, it will waste much time and is a tedious work to process those received forms by human labor. To develop a system to recognize automatically those handwritten numerals and store the result into a diskfile for the fax machine becomes more and more important today. This is the main topic studied in this paper. An assumption is made by us that the order numbers are written horizontally, and are not touched with each other. When the fax machine receives an entire line of order numbers, the proposed system will segment those numbers individually and extract 80 features for each segmented character. The result is then send to the recognition process. At the same time the fax machine receives continuously the next entire line of order numbers and the segmentation process proceeds. A self-organizing map neural network (SOM) is utilized in this study to recognize the input numerals. To promote the recognition rate, we propose two decision-typed SOM neural networks. We also develop a genetic algorithm to reduce the number of features extracted from each segmented character. Experimental results will reveal the superiority of the proposed methods.

並列關鍵字

Handwritten Numeral Recognition ； Character Segmentation ； Neural Network ； Genetic Algorithm

參考文獻

[1] Zheru Chi, Jing Wu and Hong Yan , “Handwritten Numeral Recognition using Self-Organizing Maps and Fuzzy Rules,” Pattern Recognition, vol. 28, No. 1, pp.59-66, 1995

[2] Y. J. Kim and S. W. Lee, “Off-line Recognition of Unconstrained Handwritten Digits using Multilayer Backpropagation Neural Network Combined with Genetic Algorithm,” (in Korean), in Proc. 6th Wkshp. Image Processing Understanding, PP. 186-193, 1994.

[3] Seong-Whan Lee, Chang-Hun Kim, Hong Ma and Yuan Y. Tang, “Multiresolution Recognition of Unconstrained Handwritten Numerals with Wavelet Transform and Multilayer Cluster Neural Network ,” Pattern Recognition, vol. 29, No. 12, pp.1953-1961, 1996.

[4] Zheru Chi, Mark Suters and Hong Yan , “Handwritten Digit Recognition using Combined Fuzzy Rules and Markov Chains,” Pattern Recognition, vol. 29, No. 11, pp.1821-1833, 1996

[5] Sung-Bae Cho, “Neural-Network Classifiers for Recognizing Totally Unconstrained Handwritten Numerals,” IEEE Trans. on Neural Networks. Vol. 8, NO.1, PP.43-53, 1997.