HMM BASED ONLINE HANDWRITING RECOGNITION WITHOUT
TRAINING
ABSTRACT
The basic HMM (Hidden Markov
Models) technique is regarded as the best algorithm for handwriting
recognition. But as all the systems it requires extensive training that can be
reduced to zero. In this project the HMM technique is given a boost by reducing
the training by using system font in a skewed
from 135 degree to 45 degree accommodating most of the cursive writing
and normal writing in it. This also enables us to identify any script just by
getting its system font and using it in the skew form as the training set. With
an variation of 1 degree each the normal training set of English will be 2430
per font. With 10 fonts in the database the set reaches a massive 24300
individual letters leading to a accuracy over 98%. This is the new method of
handwriting recognition that simplifies recognition and perhaps can generalize
it.
No comments:
Post a Comment