Recently, the problem of zoning design has been considered as an optimization problem and the optimal zoning is found as the one which minimizes the value of the cost function associated to the classification. For the purpose, well-suited zoning representation techniques based on Voronoi Diagrams have been proposed and effective real-coded genetic algorithms have been used for optimization. In this paper, starts from the consideration that whatever zoning method is considered, the role of feature membership function is crucial, since it determines the influence of a feature to each zone of the zoning method. Thus, in the paper the role of feature membership functions in Voronoi-based zoning methods is investigated. For the purpose, abstract-level, ranked-level and measurement-level membership functions are considered and their effectiveness is estimated under different Voronoi-based zoning methods. The experimental tests, carried out in the field of hand-written numeral recognition, show that the best results are obtained when specific measurement-level membership functions are used. © Springer-Verlag 2009.
CITATION STYLE
Impedovo, S., Ferrante, A., Modugno, R., & Pirlo, G. (2009). Feature membership functions in Voronoi-based zoning. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 5883 LNAI, pp. 202–211). https://doi.org/10.1007/978-3-642-10291-2_21
Mendeley helps you to discover research relevant for your work.