Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

 

OFF-LINE THAI HANDWRITTEN CHARACTER RECOGNITION USING HEURISTIC RULES AND NEURAL NETWORK

 

TITLE OFF-LINE THAI HANDWRITTEN CHARACTER RECOGNITION USING HEURISTIC RULES AND NEURAL NETWORK
AUTHOR YINGYOT IMPRASERT
DEGREE MASTER OF SCIENCE PROGRAMME IN COMPUTER SCIENCE
FACULTY FACULTY OF SCIENCE
ADVISOR JARERNSRI L. MITRPANONT
CO-ADVISOR SUKANYA PHONGSUPHAP
 
ABSTRACT
This research is the development of an off-line Thai handwritten character recognition system that is proposed to improve on two processes of the existing system that features extraction and recognition processes. The research framework consists of four processes: preprocessing, a feature extraction process, recognition and post-processing. The proposed functions could be classified into two components: 1) Feature extraction enhancement is used for improving the feature extraction process which consists of two sub-functions that are an additional feature conflict resolution rule and a specialized neural network-based zigzag extraction. These proposed functions are designed to refine the conflict feature and recognize zigzag patterns respectively. 2) Neural network-based recognition is the recognition process that uses a neural network as a recognizer. The neural network is used for learning Thai handwritten characters and then the trained network will be used for recognizing later. Specifically, the neural network can improve the recognition rate which can handle various styles of writing. To evaluate the proposed system, the proposed functions are developed for testing with the scanned documents. There are 10 writers obtained from previous work and gathered documents of new 41 writers, from which all of them contain 103,761 characters. According to the proposed functions, the evaluation processes are organized into four parts that are 1) The result of the existing system 2) The result of the additional feature conflict resolution rule 3) The result of the specialized neural network-based zigzag extraction 4) The result of the neural network-based recognition. Ultimately, the result could be summarized as the following: the feature extraction rate of the additional feature conflict resolution rule could achieve 87.85% (increased 2.13%), the feature extraction rate of the specialized neural network-based zigzag extraction could achieve 90.48% (increased 47.9%) and the recognition rate of the neural network-based recognition which is combined both of the two proposed feature extraction functions could achieve 92.57% (increased 9.54%).
KEYWORD NEURAL NETWORK BASED RECOGNITION / SPECIALIZED NEURAL NETWORK BASED ZIGZAG EXTRACTION

 

 

Go to Top

 

ICT Building, Mahidol University, 999 Phuttamonthon 4 Road, Salaya, Nakhonpathom 73170 Tel. +66 02 441-0909 Fax. +66 02 849-6099
Mahidol University Computing Center, The Faculty of ICT, Mahidol University , Rama 6 Road, Rajathevi, Bangkok 10400 Tel. +66 02 354-4333 Fax. +66 02 354-7333