Understanding variable shapes in 3-D

In this paper, a mathematical methodology for shape in 3-D is presented and applied to certain tasks related to the understanding of variable shape. The approach is Bayesian in nature. The methodology is based on that of a deformable template. A prior measure on polyhedral shapes is induced from a matrix-valued Gauss Markov random field on the edge graph of a polyhedral template. Stochastic relaxation is performed to obtain realizations from the posterior measure.