A Method of Generating Feature Graph for Handwritten Character Recognition of Japanese Historical Documents

In this paper, we propose a method of generating feature graphs for recognition of handwritten characters in Japanese historical documents. The feature graph represents structure of a handwritten character. In our method, the feature graph is generated as follows: (1) a rough graph is generated by thinning, clustering and other processes; (2) connections between vertices are retrieved with edge trace; (3) the curve vertices, which express curved strokes of a character, are added into the graph; (4) relative locations between vertices are calculated.