About implicit and explicit shape representation

We present a composite analysis of shapes based on form and features. We discuss how form and features are two facets of object representation and how similarity measures are used to understand the relation between two objects' images. We present a novel approach to approximate a shape that can still make use of Procrustes distance, leading to a relaxed notion of similarity measure. We introduce also a study on the similarity measures for non-parametric kernel densities. Finally we briefly discuss how these distance measures can be combined and represented into a Bayesian network, to learn the parameters of the defined similarity function. © 2006 Springer-Verlag Berlin/Heidelberg.