Evaluation of HMM based on-line handwritten character recognition with subcharacter model units for the visually impaired

摘要

This paper describes a stroke-based HMM for on-line hand-written character recognition for the visually impaired persons. In order to find the typical handwriting characteristics of visually impaired persons, three kinds of hand-written character databases have been formed. Out of them, the first one consists of 1247 characters per writer and it has been collected from visually disabled subjects, and the remaining two sets were collected from subjects with and without visual feedback. The datasets are analyzed and experiments have been performed by considering the factors that deteriorate the quality of handwriting. The experimental results show that compared to the handwriting of subjects with no visual feedback, visually impaired subjects handwriting characteristics effect the recognition result to a large extent. However, if given enough training, a visually impaired subject's writing quality is expected to reach almost the same level as it is with a person with visual feedback. Therefore the typical features of a visually impaired subject's handwriting is negligible and the proposed method is expected to work for both of visually impaired persons and persons with visual feedback.