RÉSUMÉ. Dans ce travail nous proposons de coupler la géométrie Riemannienne avec les techniques d'apprentissage pour une biométrie faciale 3D efficace et robuste aux changements d'expressions. Nous représentons localement la forme des surfaces faciales par des collections de courbes 3D. Nous appliquons des techniques d'apprentissage afin de déterminer les courbes les plus pertinentes à la reconnaissance d'identité. Les résultats obtenus sur FRGC v2 confirment
Extended abstractBiometric recognition aims to use behavioral and/or physiological characteristics of people to recognize them or to verify their identities. While some biometric modalities, such as fingerprints and iris, have already reached very high level of accuracy, they have a limited use in non-cooperative scenarios. On the other hand, the less-intrusive modalities like the face and gait have not reached the desired levels of accuracy. Since face recognition is contact-less and less intrusive, it has emerged as a more attractive and natural biometric for security applications. Unfortunately, the 2D-based face recognition technologies still face difficult challenges such as changes in illumination conditions, pose variations, occlusions, and facial expressions. In the last few years, face recognition using the shape of face surface has become a major area of research due to its theoretical robustness to challenges such as illumination and pose. Several approaches have been proposed and applied to deal with deformations caused by changes in facial expressions. In this paper, we focus on approaches that either use curve-based representations for faces or use a feature-selection technique to optimize recognition rates: (a) Curve-based approaches and (b) Feature selection-based approaches.The state-of-the-art techniques seek to analyze variability caused by facial deformations and propose methods that are robust to such shape variations. Achieving good performances in automatic 3D face recognition is an important issue when developing intelligent systems. In this paper we propose a fully automatic and unified framework for face recognition by representing a 3D facial surface by a collection of two types of curves: radial curves and iso-level curves. Furthermore, to improve performance of our identity recognition approach, we propose a geometric feature-selection approach that selects the most relevant curves by using the well-known Adaboost algorithm.The proposed framework combines machine learning techniques (Boosting) and Riemannian geometry-based shape analysis to select relevant facial curves extracted from 3D facial surfaces. The resulting set of curves provides a compact signature of 3D face, which significantly reduces the computational cost and the storage requirements for face recognition. After a 3D scan acquisition and preprocessing procedure (see Section 3) in order to extract the face area and correct some imperfections such as hole filling and spikes removing, we extract both radial and iso-level facial curves from the 3D surface. Then, acco...