A continuous shape prior for MRF-based segmentation

Abstract

The competition between discrete (MRF based) and continuous (PDE based) formulations has a very long history, especially in context of segmentation. Obviously, both have their advantages and drawbacks. Therefore the choice of a discrete or continuous framework is often driven by a particular application or (even more often) by personal preferences of a particular researcher. In this work we present a model for binary segmentation, where discrete and continuous parts are combined in a well founded and simple way. We discuss the properties of the proposed model, give a basic inference algorithm and validate it on a benchmark database. © 2013 Springer-Verlag.