Unconstrained handwritten numeral recognition using fuzzy rule-based classifier

The recognition of handwritten characters by computer has been a topic of intensive research for many years. One important branch is the handwritten numeral recognition that has many practical applications; such as mail sorting and bank check reading. Current research approaches generally fall into two categories: global analysis and structure analysis. A human can recognize an unconstrained handwritten numeral easily by simple rule-based judgment. In the paper, the variation in human writing of numerals is treated as sources of fuzziness in the numerals. Fuzzy sets and fuzzy logic are used as bases for representation of fuzzy numerals and for recognition. Only three features, fuzzy orientation, fuzzy curvature and length weight are used as the main features for recognition. This has greatly reduced the complexity of the recognition system. A rule-based fuzzy model is developed as the classifier without any training and defuzzification procedures. The system design and testing detail is discussed in the paper.