Adaptive SEJONG-NET for On-line Hangul Recognition

In this paper, a revised SEJONG-NET with adaptability for recognizing transformed on-line Hangul pattern is proposed. It is based on the structural characteristics of Hangul and the hypotheses on the processes of human Hangul recognition. Unlike the existing SEJONG-NET, the proposed model extracts the information about the orientation and the curvature of strokes in the lower layers. In the higher layers, it represents the information on graphemes as conceptual graphs, and detects a particular grapheme using the conceptual graph. The conceptual graph is composed of two kinds of nodes, concept nodes and relation nodes. The concept node has the orientation and curvature information, and the relation node describes the positional relations between two concept nodes. Through the computer simulations, we showed that the adaptive SEJONG-NET could recognize the transformed Hangul patterns after training the basic grapheme patterns, and also that untrained or severely deformed patterns can be efficiently recognized by creating a new graph or adjusting the values of the existing conceptual graphs. Hence, SEJONG-NET with adaptability can be considered as an efficient model for recognizing Hangul patterns with transformation.