Recovering the shape of a point cloud in the plane (Q1945054): Difference between revisions

Let \(X_1, \dotsc, X_n\) be an i.i.d.-sample from a distribution with compact support \(S \subset \mathbb{R}^2\). The aim is to estimate the set \(S\) from the sample. In the case where \(S\) is convex, the natural estimator is the convex hull of the sample points. This paper is concerned with estimation of \(S\) under the weaker assumption of \(\alpha\)-convexity, which means that every point of \(\mathbb{R}^2 \setminus S\) is contained in some open ball of radius \(\alpha\) that does not intersect \(S\). Given an \(\alpha\)-convex support \(S\), a natural estimator is now the \(\alpha\)-convex hull of the sample, i.e., the smallest \(\alpha\)-convex set containing \(\{X_1, \dotsc, X_n \}\). As \(\alpha\) may be unknown, the authors take a sequence of positive numbers \(r_n \to 0\) and estimate \(S\) by \(S_n\), the \(r_n\)-convex hull of \(X_1, \dotsc, X_n\). Under additional assumptions on \(S\), the law of the \(X_i\), and the sequence \(r_n\), the authors study the rate of convergence of \(S_n\) to \(S\) in a suitable sense.