Skip to Main content Skip to Navigation
New interface
Theses

Statistical Computing On Manifolds for 3D Face Analysis and Recognition

Hassen Drira 1, 2 
1 LIFL - FOX MIIRE
LIFL - Laboratoire d'Informatique Fondamentale de Lille
Abstract : We propose, in this thesis, a unified Riemannian framework for comparing, deforming, averaging and hierarchically organizing facial surfaces. This framework is applied within the 3D face recognition problem where facial expressions, pose variations, and occlusions are the main challenges of this topic. The facial surfaces are represented by collections of level curves and radial ones. The set of closed curves (level curves) constitute an infinite dimensional sub-manifold and is used to represent the nasal region, the most stable part of the face. The facial surface is represented by an indexed collection of radial curves. In this case, the calculus is simpler and the space of open curves shape is simply the hypersphere of Hilbert space. The comparison in this shape space is done via an "elastic" metric in order to handle non-isometric deformations of facial surfaces. We propose algorithms for computing means and eigenvectors in these nonlinear manifolds and hence algorithms for estimation of missing parts of 3D facial surfaces. Comparison with competitor approaches using a common experimental setting on the FRGCv2, GAVAB, BOSPHORUS databases, shows that our solution is able to obtain, and outperform in some scenarios, the state-of-the-art results.
Complete list of metadata

Cited literature [72 references]  Display  Hide  Download

https://theses.hal.science/tel-00728009
Contributor : Hassen Drira Connect in order to contact the contributor
Submitted on : Tuesday, September 4, 2012 - 4:21:29 PM
Last modification on : Friday, November 25, 2022 - 7:56:19 PM
Long-term archiving on: : Wednesday, December 5, 2012 - 10:17:38 AM

Identifiers

  • HAL Id : tel-00728009, version 1

Citation

Hassen Drira. Statistical Computing On Manifolds for 3D Face Analysis and Recognition. Computer Vision and Pattern Recognition [cs.CV]. Université des Sciences et Technologie de Lille - Lille I, 2011. English. ⟨NNT : ⟩. ⟨tel-00728009⟩

Share

Metrics

Record views

250

Files downloads

354