Preserving geometric properties in reconstructing regions from internal and nearby points (Q419376)

Let \(R\) be a region in the Euclidean space \({\mathbb R}^k\). The author considers several approaches to reconstructing \(R\) from a given set of sample points \(S\). In particular, the local convex hull constructor of radius \(\alpha\) and the \(\alpha\)-ball constructor are discussed. In the former, the reconstruction \(F(S)\) is the union of all convex hulls of subsets of \(S\) whose radius of is at most \(\alpha\). In the latter, \(F(S)\) is the union of all closed balls of radius \(\alpha\) with center \({\mathbf x}\in S\). These two methods of reconstruction are generalized in the notion of a locally-based region constructor. It is shown that if \(R\) is a bounded regular region \(R\), then there exists \(\alpha>0\) such that for any locally-based region constructor \(F\) of maximal radius \(\alpha\), the dual Hausdorff distance between \(R\) and \(F(S)\) (that is, the greater of the Hausdorff distance between \(R\), \(F(S)\) and that of the closure of their complements) is small, as is the Hausdorff distance between their respective boundaries, provided that the Hausdorff distance between \(R\) and \(S\) is small enough. In addition, for the local convex hull constructor of radius \(\alpha\) and the \(\alpha\)-ball constructor, the author shows that if \(R\) has a reasonable boundary, then \(F(S)\) is \(\epsilon\)-similar to \(R\) whenever \(\alpha\) and the Hausdorff distance between \(R\) and \(S\) are small enough. That is, there is a homeomorphism \(\Gamma\) on \({\mathbb R}^k\) such that \(F(S)=\Gamma(R)\) and the distance between any point \({\mathbf p}\) and its image \(\Gamma({\mathbf p})\) is at most \(\epsilon\). Moreover, for a given angle \(\phi\), \(F(S)\) can be taken to be an \((\epsilon,\phi)\)-approximation in tangent of \(R\): if \({\mathbf q}\) is a point on the boundary of \(F(S)\), then the angle between the normal vectors at \({\mathbf q}\) and \(\Gamma({\mathbf q})\) is at most \(\phi\).