Writer independent on-line handwriting recognition using an HMM approach

In this paper we describe a Hidden Markov Model(HMM) based writer independe nt handwriting recognition system. A combination of signal normalization pr eprocessing and the use of invariant features makes the system robust with respect to variability among different writers as well as different writing environments and ink collection mechanisms. A combination of point oriente d and stroke oriented features yields improved accuracy. Language modeling constrains the hypothesis space to manageable levels in most cases. In addi tion a two-pass N-best approach is taken for large vocabularies. We report experimental results for both character and word recognition on several UNI PEN datasets, which are standard datasets of English text collected from ar ound the world. (C) 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.