A hybrid n-tuple neuro-fuzzy classifier for handwritten numerals recognition
RAIDA AL-ALAWI · 2005
A hybrid neuro-fuzzy system applied to the classification of handwritten numerals is presented. The system combines the advantages of the n-tuple sampling technique and fuzzy inference system. The n-tuple unit is used as a preprocessing unit for extracting the feature vector from the input pattern. The outputs of the n-tuple unit are fed to a fuzzy inference unit that applies a set of fuzzy rules on the feature vectors and aggregates them to generate its classification response. The classification accuracy of the n-tuple neuro-fuzzy system and the classical n-tuple classifier is compared using handwritten numerals from NIST database. The n-tuple neuro-fuzzy classifier achieves an accuracy of 98.5% on classifying unseen numerals.