Set estimation under convexity type assumptions

The problem of estimating a set S from a random sample of points taken within S is considered. It is assumed that S is r-convex, which means that a ball of radius r can go around from outside the set boundary. Under this assumption, the r-convex hull of the sample is a natural estimator of S. We obtain convergence rates for this estimator under both the distance in measure and the Hausdorff metric between sets. It is also proved that the boundary of the estimator consistently estimates the boundary of S, in Hausdorff's sense. (C) 2007 Elsevier Masson SAS. All rights reserved.