EXPERIMENTAL-STUDY OF A NOVEL NEURO-FUZZY SYSTEM FOR ONLINE HANDWRITTEN UNIPEN DIGIT RECOGNITION

This paper presents an on-line hand-printed character recognition syst em,tested on datasets produced by the UNIPEN project, thus ensuring su fficient dataset size, author-independence and a capacity for objectiv e benchmarking. New preprocessing and segmentation methods are propose d in order to derive a sequence of strokes for each character, followi ng suggestions of biological models for handwriting. Variants of a nov el neuro-fuzzy system, FasArt (Fuzzy Adaptive System ART-based), are u sed for both clustering and classification. The first task assesses th e quality of segmentation and feature extraction techniques, together with an analysis of Shannon entropy. Experimental results for classifi cation of the train_r01_v02 UNIPEN dataset show real-time performance and a recognition rate of over 85%, exceeding slightly Fuzzy ARTMAP pe rformance, and 5% inferior to the rate achieved by humans. (C) 1998 El sevier Science B.V. All rights reserved.