Many of man-machine communications have been made by means of the keyboard. The interface is not well-suited to users who have not mastered the keyboard and if, in addition, the number of strokes becomes large (as in the Chinese of Hangeul characters), the communication becomes cumbersome and inefficient. For these reasons, on-line handwriting recognition systems have been developed.
In general, character recognition system consists of three parts : preprocessing, stroke classification and character recognition. In this thesis, the scheme of the stroke classification based on the vector representation and of the character recognition using the dictionary lookup method are proposed. Handwritten characters are made of strokes, and each stroke is represented by a set of vectors. A set of actual vectors represents the shape of a stroke while a set of virtual vectors represents the relations among the strokes are used in stroke matching. After the strokes matching process, characters are sequentially recognized in the order of choseong, jungseong and jongseong.
Experiments are done with 8 sets of 506 Hangeul characters written by 8 men. Among the sample data sets, 6 data sets are used to construct jaso models and to find the possibility of the inter-jaso relations. the system achieved 93% recognition accuracy for the training data, and 82% for the test data. Through the experiments, the results show that increasing learning sets makes the similarity between jaso-model larger.