A character recognition system with learning dictionary for handwritten drawings


This paper discusses the character recognition system used in the automatic input of drawings. ered is of small size, 2 to 5 mm, which is sampled with a low resolution such as 5 lines/mm. The recognition is performed by a kind of pattern matching. The input pattern is superposed on the dictionary pattern. The distance distortion measure is introduced which represents the degree of mismatch with the closest pixels as the corresponding pair. To reflect the structural feature of the character, the weighted distortion measure is proposed which considers the environment of the corresponding pixels. By this elaboration, the misrecognition rate was decreased by 45 percent. the problem arising from representing the two-dimensional pattern distortion by a single number, dictionary patterns are increased. The generation and the structure of the dictionary are discussed. A hierarchical dictionary with learning pointer is constructed which records the mistakes made in the course of training. The feature of the method is that mistakes made in the past can be utilized and the processing time does not increase so rapidly in proportion to the number of dictionary patterns. Approximately 30,000 characters were learned, and the practical recognition performance was verified through the recognition experiment for the drawings. The character consid-

DOI: 10.1002/scj.4690180207

10 Figures and Tables

Cite this paper

@article{Ishii1987ACR, title={A character recognition system with learning dictionary for handwritten drawings}, author={Mitsuo Ishii and Michiko Iwasaki and Masahiro Yamada}, journal={Systems and Computers in Japan}, year={1987}, volume={18}, pages={65-76} }