On-line cursive script recognition: a system that adapts to an unknown user

This paper presents a system dedicated to the recognition of English cursive words drawn on a digitizing tablet. This system uses an analytical approach in the sense that it tries to localise the letters of the word to be recognised due to a set of letter prototypes. Its main originality is its ability to upgrade its prototype set at run-time. This is done based on a recognition module that produces an accurate letter segmentation of the words it identifies. For a given user, this adaptation ensures a very good performance after the recognition of a few hundred words. Four series of experiments demonstrate the efficiency of this new approach.