Rejection criteria and pairwise discrimination of handwritten numerals based on structural features

This paper presents a new structural method for the recognition of handwrit ten numerals. Contour shape features, such as convex arcs, concave arcs, li ne segments, end-point arcs and holes from the contours of numeric characte rs, are used to describe numeral characters. A new method of measuring the similarity between a sample and class is proposed. A two stage recognition methodology is also presented, in which two rejection criteria are introduc ed. In the first stage of recognition, an input sample is given an identity or categorised as either first class or second class rejection, based on s imilarity measures between the input sample and each of the ten numeral cla sses. In the second stage of recognition, strategies are introduced to modi fy the structural description of the input sample if it is in first class r ejection and a classifier focused on pairwise discrimination is applied if the input sample is in second class rejection. Experimental results indicat e that the overall performance of the proposed method compares favourably w ith chose achieved by other methods found in the literature.