Variational segmentation with shape priors

Abstract

We discuss the design of shape priors for variational region-based segmentation. By means of two different approaches, we elucidate the critical design issues involved: representation of shape, use of perceptually plausible dissimilarity measures, Euclidean embedding of shapes, learning of shape appearance from examples, combining shape priors and variational approaches to segmentation. The overall approach enables the appearance-based segmentation of views of 3D objects, without the use of 3D models. © 2006 Springer Science+Business Media, Inc.