Recognition of unconstrained handwritten numerals using crossing features

This paper presents a method of recognizing unconstrained handwritten numerals. For feature extraction, a method called crossing feature coding is proposed. The method scans through the rows and columns to obtain the horizontal and vertical crossing times of a character. The method is fast in extracting features and it can provide a large amount of information for classification. Based on the horizontal and vertical crossing feature codes and some other feature such as stroke length and slope, and density, a number of classifying rules are proposed. For classification a self-correcting classification tree is designed. A practical recognition and mail sorting system is set up to test the effectiveness of the recognition method