An integrated neural architecture for recognition, correction and storage of handwritten technical documents

A neural architecture for processing of on-line handwritten documents based on adaptive resonance theory (ART) models is proposed in this paper. The first architecture, on which a laboratory prototype of a mathematical editor was built, uses ART-2 and ARTMAP modules for the successive classification of symbol components, and symbols respectively. A STORE model is necessary in order to obtain a spatial pattern for the symbol from the sequence of component nodes. An enhanced architecture, based on fuzzy ART and fuzzy ARTMAP modules, is also presented, since it can exploit in a better way the information provided by the supervision in the learning phase. Finally, an overview of a complete integrated architecture is given, that incorporates the information provided by a syllable based dictionary, as well as the process to obtain an office document architecture (ODA) description of the on-line produced technical document