Chinese Handwriting Recognition An Algorithmic Perspective

Author : Ho-Ting Wong
Publisher :
ISBN 13 : 9781374709737
Total Pages : 0 pages
Book Rating : 4.7/5 (97 download)

DOWNLOAD NOW!

Download or read book Hand-Written Chinese Character Recognition by First and Second Order Hidden Markov Models and Radical Modeling written by Ho-Ting Wong and published by . This book was released on 2017-01-27 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: This dissertation, "Hand-written Chinese Character Recognition by First and Second Order Hidden Markov Models and Radical Modeling" by Ho-ting, Wong, 黃浩霆, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled "Hand-written Chinese Character Recognition by First and Second Order Hidden Markov Models and Radical Modeling" Submitted by Wong Ho Ting for the degree of Master of Philosophy at The University of Hong Kong in December 2003 Numerous handwritten Chinese character recognition systems are available on the market at present. Most of them are not robust enough, and handwritten Chinese character recognition remains a common research topic in pattern recognition. This research project exploits the fact that most Chinese characters can be broken down into simpler units (radicals). We hypothesize that using information from radicals will improve the recognition rate, and propose a recognition system deploying radical modeling. We also show that the recognition rate can be improved by using second order Hidden Markov Model (HMM) classifiers instead of first order HMM. A method for improving the rate of recognition of Chinese characters by modeling radicals instead of the whole character is presented. Because a radical is necessarily more primitive than the character in which it occurs, the complexity of the character models can be significantly reduced. The application of this concept requires an effective radical extraction scheme, and a modified version of Han's radical extraction method is presented. However, because we believe that the concept of radical modeling should be used only when it is needed, we introduce a criterion into the recognition system that determines when to apply the concept. We also compare the performance of second order HMM classifiers against first order HMM classifiers, due to their greater descriptive power. Experiments were conducted on the ETL8B2 character set, and a 0.43% improvement was obtained by introducing the concept of radical modeling in a recognition system with a first order HMM fine-classifier and a Bayes pre-classifier. This is an encouraging result because this 0.43% represented about one third of all characters which were wrongly recognized by the recognition system without radical modeling. If a comparison is made only with a first order HMM classifier, a 0.52% improvement can be observed. On the other hand, although second order HMM did not show a constant improvement from the experiment results, the improvement in the case of codebook size equal to 64 suggests that constant improvement may be observed by having enough training samples or performing a KL transform in the training phase. In conclusion, we have proposed a method for improving the recognition rate of handwritten Chinese characters by using the concept of radical modeling. We applied this concept to certain appropriate character samples, using a modified version of Han's radical extraction method in the recognition system. We also compared the performance of a second order HMM classifier with a typical first order HMM classifier. Encouraging results were observed, and the results indicate that a combination of radical modeling and use of second order HMM could help to improve the rate of recognition of handwritten Chinese characters. DOI: 10.5353/th_b2777086 Subjects: Pattern recognition systems Chinese characters - Data processing Markov processes