Shape Registration and Establishing Correspondences in Implicit Spaces

Sharon Huang
Department of Computer Science, Lehigh University

Monday, November 6, 2:00PM
Babbio Center, Room 110
Stevens Institute of Technology
 

Abstract


This talk explores shape representation and registration through implicit distance functions. A shape of interest in its implicit form corresponds to the zero level-set of a higher dimensional distance function. In certain applications such as shape registration, which aims to recover a transformation that brings a source shape to achieve high spatial correspondence with a target shape, the implicit representation has advantages because it provides additional support to the registration process and requires matching of not only the shapes but also their clones that are positioned coherently in the embedding image/volume space.

This talk will present a new variational and statistical approach to shape registration using implicit representation. The Mutual Information criterion supports various transformation models and is optimized to perform global registration (i.e. alignment). Then a B-spline based Incremental Free Form Deformations (IFFD) model is used to minimize a Sum-of-Squared-Differences (SSD) measure and further recover a dense local non-rigid registration field. The framework deals with shapes of arbitrary dimension and topology (multiple parts, closed/open), it preserves shape topology during local deformation and produces local registration fields that are smooth, continuous and establish one-to-one correspondences. The potential of the framework is shown on two applications -- statistical modeling of anatomical structures, and 3D/4D facial expression tracking. Its performance is also compared with that of several other shape registration algorithms.