“…When the law is unknown, nonparametric methods can be used that do not make a priori assumptions about distribution. These are subdivided into two classes: first, scalebased methods for which equal width (Chu, 1991;Mostardi, 1990), equal-height (Liou and Yao, 1977;Muralikrishna and DeWitt, 1988;Piatetsky-Shapiro and Connell, 1984) and variable-width (Merrett and Otoo, 1979;Muthuswamy and Kerschberg, 1985;Whang et al, 1994) histograms are well-known examples, and second, analytic methods which approximate the actual data distribution using a finite series of analytic functions (Lefons, 1988;Lefons et al, 1983Lefons et al, , 1995Sun et al, 1993). Guillaud and Chassery (1992) used a pattern recognition approach to interpret and analyze histograms which were defined in a new vectorial space determined by Legendre polynomials.…”