A Comparison of Recognition for Off-line Myanmar Handwriting and Printed Characters

We propose a system to recognize off-line Myanmar handwriting and printed characters, the language used by the majority of Myanmar. This system is based on the use of discrete hidden Markov models. Myanmar handwriting and printed characters or words were chosen for the study. After preprocessing step, the characters are auto-segmented using recursive algorithm as sequences of connected neighbors along lines and curves. The unknown Myanmar characters and words are first pre-classified into one of known character groups, based on the structural properties of the text line. The hidden Markov models classifier is then proposed for the final recognition. All the characters were written by the different writers on a preformatted paper. The method recognizes the Myanmar handwriting in print style. The system was trained and tested Myanmar characters images. All the characters were written by 5 different writers on a preformatted paper. A comparison results have shown 92.1% recognition rate for the handwriting and 97% recognition rate for the printed characters