Some properties of stochastic labeling procedures

Abstract

In the paper an optimization model for stochastic labeling is formulated and the first-order Kuhn-Tucker conditions of the model are derived. The nonzero fixed points of the relaxation scheme by Rosenfeld, Hummel, and Zucker are then characterized in terms of these conditions. Further a local convergence result is presented and proved and the use of the relaxation scheme for optimization is discussed.