Multigrid Convergent Curvature Estimator
Abstract
We propose in this paper an estimator of derivative and curvature of discrete curves. Based on adaptive convolution that preserves contour, we use local geometrical information as the heat kernel to con-volve with a discrete curve and give estimation of its geometrical parameters. We recover on regular part of the curve the classical convolution based on gaussian kernel. We study the bounded error of our approach for first and second order derivative and we discuss about the multigrid convergence.