47
11

Estimating perimeter using graph cuts

Abstract

We investigate the estimation of the perimeter of a set by a graph cut of a random geometric graph. For ΩD=(0,1)d\Omega \subset D = (0,1)^d, with d2d \geq 2, we are given nn random i.i.d. points on DD whose membership in Ω\Omega is known. We consider the sample as a random geometric graph with connection distance ε>0\varepsilon>0. We estimate the perimeter of Ω\Omega (relative to DD) by the, appropriately rescaled, graph cut between the vertices in Ω\Omega and the vertices in D\ΩD \backslash \Omega. We obtain bias and variance estimates on the error, which are optimal in scaling with respect to nn and ε\varepsilon. We consider two scaling regimes: the dense (when the average degree of the vertices goes to \infty) and the sparse one (when the degree goes to 00). In the dense regime there is a crossover in the nature of approximation at dimension d=5d=5: we show that in low dimensions d=2,3,4d=2,3,4 one can obtain confidence intervals for the approximation error, while in higher dimensions one can only obtain error estimates for testing the hypothesis that the perimeter is less than a given number.

View on arXiv
Comments on this paper