Multi-feature Extraction for Printed Thai Character Recognition

This paper presents a simplified printed Thai character recognition system using multiple feature extraction and character classification. Three relevant information extracted from a set of training character images are the direction of each character’s contour, the density of character body and character peripheral information. This set of features is used as reference for classifying unknown input characters. Based on Euclidean space model, the category of the reference vector yielding the minimum distance is assigned to the input character pattern. From the experiments, the recognizing speed is 5 character images per second with 97.44% correctness. The performance of recognition can be improved gradually in strengthening the robustness and lowering the error recognition rate by simply training and maintaining the knowledge by users.