Covered data structures I

The algorithm

Institut für Astronomie der Leopold–Franzens–Universität Innsbruck, Technikerstraße 25, A–6020 Innsbruck, Austria http://astro.uibk.ac.at

Send offprint request to: S. Kimeswenger

Received: 13 September 1996
Accepted: 26 May 1997

Abstract

Many  algorithms  separating  or  detecting groups of similiar objects (for example the extraction of groups lying in a color–color–diagram) are based on two statistical methods: the Kernel Method (Silverman 1986) or the Likelihood Statistic (van der Waerden 1957). These standard methods have one or more restrictions (e.g. known number or differentiability of the groups, ). We present here a new powerful algorithm and show results worked out with artificial data sets. 
The algorithm is based on Recursive Restoration Methods (neither on the Likelihood Statistic (Sutherland & Saunders 1992) nor on the Kernel Method (De Jager et al. 1986) and allows to detect substructures in a data set, even if they are overlapped or superimposed by any kind of dominating main structure. In comparison to the other methods mentioned above there are no restrictions concerning the form and the dimension of the components lying in the data set. 
The algorithm is easy to handle and therefore opens a wide range of applications for many fields of science (see Boller et al. 1992).