On-Line Cursive Handwriting Recognition. Phase 1.

摘要

There is a widely recognized need for improved user interfaces that will make computers easier to use for people with no technical training, and improve the general efficiency and ergonomics of many computer applications where the conventional keyboard is too limiting. One approach is based on handwriting and gesture recognition, which will enable people to use common handwriting skills to communicate with computers for a wide variety of applications such as text editing, form filling, and interactive graphics. But before handwriting/gesture interfaces can become widely popular, accurate and reliable recognition of normal cursive writing is necessary. The goal of the study was to assess the technical feasibility of cursive handwriting recognition. The cursive recognition algorithm described consists of several parts: a front-end Dynamic Programming (DP) algorithm, extensions to the DP algorithm that find not only the best match but also all plausible alternatives, dictionary matching, rules-based discrimination to resolve confusions, and syntax constraints. Tests on a data base of cursive words resulted in 99.2% correct recognition in user-independent mode, and 99.9% in user-dependent mode.