Handwritten numeral recognition using a sequential classifier
L.L. Lee, Natanael Gomes · 2002
An approach for numerical character recognition involving discriminating feature extraction and neural classification is proposed. The image of an unknown numeral is firstly preprocessed in order to guarantee the extraction of good features. The preprocessing operation consists of scale normalization, image thinning, elimination of spurious segments and image dilation. Then, some discriminating features are extracted from the normalized image and used for numeral classification. The classification process is divided into two steps. In the first step, the unknown numeral classification is based on the image's topological features and the image pixel distribution. In the second step of classification, Hopfield nets are used. Experimental tests on handwritten numerals written on white paper and bank checks reveal that the recognition rates of 85% an 92.4% are achieved, respectively.