Tuning between Exponential Functions and Zones for Membership Functions Selection in Voronoi-based Zoning for Handwritten Character Recognition

In Handwritten Character Recognition, zoning is rightly considered as one of the most effective feature extraction techniques. In the past, many zoning methods have been proposed, based on static and dynamic zoning design strategies. Notwithstanding, little attention has been paid so far to the role of function-zone membership functions, that define the way in which a feature influences different zones of the pattern. In this paper the effectiveness of membership functions for zoning-based classification is investigated. For the purpose, a useful representation of zoning methods based on Voronoi Diagram is adopted and several membership functions are considered, according to abstract–, ranked- and measurement-levels strategies. Furthermore, a new class of membership functions with adaptive capabilities is introduced and a real-coded genetic algorithm is proposed to determine both the optimal zoning and the adaptive membership functions most profitable for a given classification problem. The experimental tests, carried out in the field of handwritten digit recognition, show the superiority of adaptive membership functions compared to traditional functions, whatever zoning method is used.