標題: | 貝氏決策型神經網路於手寫中文字辨識之研究 The study of handwritten Chinese character recognition by Bayesian decision neural network |
作者: | 徐永煜 Xu, Yeong-Yuh 傅心家 Hsin-Chia Fu 資訊科學與工程研究所 |
關鍵字: | 貝式決策網路;複合式高斯分佈;動態調整型;Bayes decision neural network;Mixture Gaussian distribution;Dynamic allocate |
公開日期: | 1996 |
摘要: | 本論文的目的是應用貝式決策型類神經網路於手寫中文文字之辨識。本論 文所能辨識的字為教育部所選定的 5401 個常用字。論文中提到了一些貝 式決策型網路的家族成員,包括單一分佈型貝式決策網路、複合式高斯分 佈型貝式決策網路、模糊型貝式決策網路、動態調整型貝式決策網路,並 對其訓練法則有詳盡的描述。本辨識系統包括五個步驟,依序是前處理、 特徵擷取、大分類、中分類以及細分類。在前處理的部分,包括平滑化、 大小正規化、非線性正規化以及細線化。在特徵擷取流程,我們以筆畫穿 越數,帶狀黑點數及筆畫走向當做辨識特徵。總維度為92維。在大分類辨 識流程,我們採用了重疊式的群集演算法作為大分類器,以加快系統之辨 識速度。在細分類辨識流程,我們使用貝氏決策型神經網路作為主辨識器 ,進行單字之辨識。論文中亦完成了一個以貝式決策型網路實作的手寫中 文文字辨識系統。以前100人作為訓練及測試,其訓練正確率可達 97.92%以上,其測試辨識率可達82.16%。若取200人作為訓練及測試,其 訓練正確率可達 97.50%以上,其測試正確率則為 64.72%。 This thesis presents an application of Bayesian Decision Based neural networks on off-line handwritten Chinese characters recognition. Our recognition system includes preprocessing, feature extraction, coarse classification and major classification.In order to evaluate the proposed recognition system, we choose 5401 frequently used Chinese characters as our trainning and testing domain. The database of each testing and trainning sample character was created by the Computer and Communication Laboratory of Industrial Technology Research Institute.Experimental results show that, for the inside testing of the 100 personsdatabase, the recognition rate is about 97.92%. The recognition rate is about 82.16% in the outside testing in the 100 persons database. When thedatabase is changed to 200 persons, for the inside testing, the recognitionrate is about 97.50%. The recognition rate in the outside testing is about 64.72%. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#NT850392033 http://hdl.handle.net/11536/61783 |
顯示於類別: | 畢業論文 |