Combining statistical and structural approaches for handwritten character description

In this paper a new character description method, based on the combination of structural and statistical approaches, is presented. Characters are preliminarily decomposed in terms of structural primitives (circular arcs) and successively described in terms of statistical features (geometric moments). The obtained description is much more stable and yields significant improvements in classification performance: its effectiveness has been demonstrated by comparing the recognition results obtained by applying the geometric moments directly on the character bit maps and, as proposed, on the character decomposition in circular arcs. Absolute and relative performance is significant especially for particularly critical cases. Novel recurrent formulae for evaluating in a closed form the moments of objects represented in terms of circular arcs are also introduced; experimental results reveal a significant reduction of the time needed for evaluating the moments.