Statistical shape models for tubular objects

Statistical shape models are able to capture the natural variability in shape that is present within a certain class of objects. The mayor challenge in building a statistical shape model is the construction of a pointwise dense correspondence between the objects in the sample, which can be established through parameterization. Current techniques usually parameterize the objects on a sphere, which works fine for sphere-like objects. However, for elongated or tubular objects the quality of the spherical parameterization degrades as well as the quality of the correspondence. In this paper, a method is proposed that finds correspondences between the objects of the statistical sample by parameterizing them on the cylinder. It is shown that the quality of these parameterizations is better and that the model built from it better represents the shape variability of the object class.