Wavelet modeling of contour deformations in Sobolev spaces for litting and tracking applications

Abstract

This paper proposes a new model for contour deformations using wavelets. This model uses Sobolev spaces to control the smoothness of the contour deformation. This formulation defines a probabilistic model that induces a prior distribution for contour deformation. Based on this distribution, the fitting problem is solved in Bayesian terms. The deformation model is also used to generate a prior dynamic model for contour evolution in time. This probabilistic model is then applied to solve the tracking problem. Computational results for several real-image problems are given for both the Kalman and condensation filters.