Using deformable surfaces to segment 3-D images and infer differential structures
Abstract
In this paper, we use a 3-D deformable model, which evolves in 3-D images, under the action of internal forces (describing some elasticity properties of the surface), and external forces attracting the surface toward some detected edgels. Our formalism leads to the minimization of an approach and a conforming finite element method to actually express the surface in e discrete basis of continuous functions. This leads to reduced computational complexity and better numerical stability. The power of the approach to segmenting 3-D images is demonstrated by a set of experimental results on various complex medical 3-D images. Another contribution of this approach is the possibility to infer easily the differential structure of the segmented surface. As we end up with an analytical description of class of the surface almost everywhere, this allows us to compute, for instance, its first and second fundamental forms. From this, one can extract a curvature primal sketch of the surface, including some intrinsic features which can be used as landmark for 3-D image interpretation?
Origin : Files produced by the author(s)