PhD Defense of Florent Brunet
Title: Contributions to Parametric Image Registration and 3D Surface Reconstruction
Abstract: This thesis deals with the modelling and the estimation of parametric functions in computer vision. Three main topics are considered and are related by a single objective: the reconstruction of an arbitrary surface from images taken in an arbitrary deformable environment. The three main topics considered in this work are:
- Range surface fitting: we consider the problem of fitting a parameteric surface such as a B-spline to a set of range data obtained with, for instance, a time-of-flight cameras. We take a particular interest in the problem of automatically determining the hyperparameters. We also propose methods to cope with discontinuities and heteroskedastic noise.
- Image registration: several problems related to image registration are considered. In particular, we propose a new method for direct image registration that does not need a region of interest. We also propose new ways of automatically selecting the hyperparameters in feature-based image registration. We finally propose a new model of deformation, the NURBS-warp, that copes with perspective effects in the imaging process.
- Reconstruction of inextensible surfaces from monocular videos: the last part of this thesis tackles the difficult problem of reconstructing a deformable surface in a monocular setup. We propose two new approaches to solve this problem ; The first one is a convex formulation of some common assumptions ; The second one is a new way of enforcing the inextensibility constraint in a parametric framework.
Advisors: Pr. Adrien Bartoli, Pr. Nassir Navab, Pr Rémy Malgouyres
Examiner: Pr. Laurent Sarry
Referees: Pr. Joachim Hornegger, Dr. Lourdes Agapito, Pr. Etienne Mémin
Location: Campus Universitaire des Cézeaux, Clermont-Ferrand, France