Shape Matching Using the Geodesic Eccentricity Transform - A Study

Abstract

This paper makes use of the continuous eccentricity transform to perform shape matching. The eccentricity transform has already been proven useful in a discrete graph-theoretic setting. We show how these ideas extend naturally to the continuous setting thus bringing a higher geometrical fidelity. The continuous eccentricity transform is used to compute multiscale descriptors for shapes. These descriptors are defined as histograms of the eccentricity transform of a scale-space representation of the shape. These multiscale descriptors are naturally invariant to euclidean motion and bending. They show promising results for shape discrimination.