On-Line Character Analysis and Recognition With Fuzzy Neural Networks
Eduardo Gómez‐Sánchez, Yannis A. Dimitriadis, M. Sánchez-Reyes Más, P. Sánchez Gracía, J.M. Izquierdo, Juan López Coronado · Intelligent Automation & Soft Computing · 2001
A new recognition system based on a neuro-fuzzy systcnn, called FasArt, is proposed in this paper. Satisfactory restfits were obtained using the train rO1 v02 UNIPEN dataset, together with a comparison with the recognition rates achieved by independent human testers. Two methods for segmenting handwritten components into strokes are proposed, with better experimental restfits for the method based on biological models of handwriting, in terms of consistency and network complexity. A systematic experimental study of different codification schemes is also described, based on Shannon entropy and clustering maps. Finally, some steps towards the construction of an allograph lexicon are shown, that exploit the generation of fuzzy-rules by FasArt architecture.