Asymptotical minimax recovery of sets with smooth boundaries (Q1896268)

We derive optimal convergence rates for estimates of sets in \(N\)- dimensional ``black-and-white'' pictures. We consider estimation of ``black'' regions under smoothness conditions. It is assumed that the boundaries of these regions have a smooth parameterisation, so that the regions belong to the Dudley classes of sets [\textit{R. M. Dudley}, J. Approximation Theory 10, 227-236 (1974; Zbl 0275.41011)]. The variety of possible sets covered by our assumptions is rather large since smooth parameterisation of the boundary does not imply that the boundary itself is smooth. Also for convex sets our approach gives the correct rate of convergence. We study the problem of estimation of sets for two different setups: edge estimation and density support estimation. The study of edge estimation is motivated by applications in image analysis. Often as a first step in image processing the original picture is transformed to a binary black- and-white picture which is then used to recover edges, that is, boundaries of black regions. Density support estimation is related to cluster analysis and quality control. The aim of this paper is not to add some new practical recipe to the variety of existing ones, but to propose, under certain idealisations, a tool for comparing the statistical properties of different estimation methods. Such a tool, in our opinion, is provided by the knowledge of the optimal convergence rates.