Improving handwritten numeral recognition using fuzzy logic

The author presents an improved method of handwritten numeral recognition using fuzzy logic. In handwritten numeral recognition, most recognition errors are found in confusing samples. To represent confusing features and improve recognition rates, he groups confusing numerals into confusion groups and builds fuzzy functions applying human knowledge. To use small and delicate features of numerals, he structurally represents a numeral as a sequence of primitive strokes and feature points. To compensate weaknesses of the structural method, he also uses a neural network method. Experimental results on collected test samples show the efficiency and robustness of the proposed method.