Exploring the Ability of Parts on Recognizing Handwriting Characters

This paper shows an experimental trial on part-based online character recognition. The purpose of the trial is to evaluate the discrimination ability of parts of a handwriting character, that is, a temporal pattern showing a character. Each part is a temporal segment of the character and the character is decomposed into a set of parts. Those parts are then represented as a “bag-of-features”, which is a histogram showing how many parts similar to a specific representative part exist in the character. This representation totally disregards the global structure of the character and thus invariant to any deformation in the global structure, such as stroke order variations. Our experimental result has revealed that, without any global structure information, such as position of each part, relative relationship between parts, and temporal order of parts, 50-60% accuracies can be still achieved for 80 elementary Chinese character classes.