On Matching Deformable Models to Images

Abstract

A crucial goal of machine vision is the computation of mathematical models describing the shapes of objects appearing in images. In this paper we formulate a broadly applicable class of models which may be viewed as deformable bodies composed of an abstract elastic material. The models are matched to images through the action of external forces which deform the bodies from their prescribed natural states and impart motions toward desired equilibrium configurations. Forces are typically derived from images, but they can also be applied interactively by the user (say, by means of a pointing device) or abstracted from physical phenomena (such as gravity).