On-line cursive script recognition using time delay neural networks and hidden Markov models.

M. Schenkel, I. Guyon, and D. Henderson.
Machine Vision and Applications, pages 215--223.

A system for the on-line recognition of cursive script is presented. As input data we use the trajectory data as recorded by a touch sensitive pad, such as the ones used by note-pad computers. The input strings consist of a time-ordered sequence of X-Y coordinates, punctuated by pen-lifts. A neural network recognizer with local connections and shared weights is used to estimate the a posteriori probabilities for characters in a word. A Hidden Markov Model with fixed parameters segments the word into characters in a way which optimizes the global word score taking a dictionary into account. A word normalization scheme for cursive and a fast but efficient dictionary search are also presented. Trained on 20k words from 59 writers and using a 25k word dictionary we reached a 89 character and 80 word recognition rate on unconstrained data of different writers.

Keywords: on-line character recognition, neural networks, segmentation, cursive script handwriting, hidden Markov models, dictionary search.

[ next paper ]