An Aerial Handwritten Character Recognition Using Moving Direction Only

Abstract

This paper describes a method to recognize and learn a character handwritten in the air. This recognition method uses the motion direction instead of positions of the device. It also doesn't use the information of pen up and down. It selects and orders character candidates with DP (dynamic programming) matching. To use this matching, at first, we analyzed standard character shapes and made penalty codes and dictionary. However, the recognition rates was not stable. Therefore, we adopted Genetic Algorithm in order to define those parameters. As a result, we get general penalty codes which doesn't depend on users and dictionary. And we improve recognition rates of bad operator by learned dictionary. But recognition rates with evaluation data by learned dictionary is not so effective compare with learned penalty codes.