An improvement of on-line Japanese character recognition system for visually disabled persons

We developed an online Japanese character entry system with handwriting input device for visually disabled persons (particularly, acquired blind persons). This system is composed of an electric tablet for writing a character, a hand controller for controlling the system condition, a recognition algorithm of handwritten characters, and a voice synthesizer which indicates a recognized character. We investigated the characteristics of Japanese character distortion written by visually disabled persons. As the result, we found that a relative position of stroke representative points (a stroke means a continuous line segment from pen-down to pen-up) in the partial pattern is stable, even if the written characters were extremely distorted. Therefore we already proposed a character recognition method based on the relative direction between two strokes (the RDS method) for characters written by visually disabled person. However, as the number of feature parameters of the RDS method are decreasing for the characters having less than 5 strokes, recognition errors are increasing. This paper describes the fusion (improvement) method by which a method based on the line segment directions in a stroke is added to the RDS method. From the experimental results, total recognition rate of a 92.5% (cumulative rates for the three best candidate of a 99.1%) are obtained from 1192 Japanese characters test for 11 persons