Project

Back to overview

Graphical Models pour l'authentification de visages

English title Graphical Models for face authentication
Applicant Marcel Sébastien
Number 119921
Funding scheme Project funding (Div. I-III)
Research institution IDIAP Institut de Recherche
Institution of higher education Idiap Research Institute - IDIAP
Main discipline Information Technology
Start/End 01.07.2008 - 30.06.2009
Approved amount 53'948.00
Show all

Keywords (5)

machine learning; graphical models; face authentication; Face Recognition; Bayesian Networks

Lay Summary (English)

Lead
Lay summary
In this project, the problem of face authentication is addressed. Face authentication consists in either accepting or rejecting the identity claim of an individual supported by its face image, hence it falls into the more generic problem of Face Recognition. Although current algorithms are quite successful on controlled conditions, performance decreases rapidly in case of unconstrained viewing conditions. Recently, authentication systems based on local features and statistical models were shown to perform well for the authentication task. However, currently used models are not appropriate to the structure of the observed data, and implicitly assume independence between the observed local facial features.We thus intend to investigate the usage of the more generic Bayesian network framework as statistical models for face authentication. Bayesian networks provide an intuitive way to represent the joint probability distribution over a set of variables. In this framework, it is then possible to model relationships between the different observations derived from a face image, and hence to relax the independence assumptions. However, finding a set of observations together with an appropriate network structure is an open research issue.Another property of Bayesian networks is their ability to handle arbitrary random variables. Hence, we intend to add auxiliary information to the system, so as to improve its accuracy. As evidenced by psychological studies, human beings use several visual clues to recognise a face. Examples of such information include the global face shape and the skin colour for instance. Moreover, it is also possible to model external sources of variations directly in the network, by simply adding appropriate nodes to the graph. As mentioned in previous studies, pose and illumination are the two major factors affecting face recognition system performance, hence robustness may be gained if such variations are accounted for.We believe that Bayesian networks are well suited to the face authentication task mainly for two reasons. First, explicit relationships between facial features can be encoded through the network structure, hence relaxing independence assumptions. Second, we believe that using auxiliary information, as well as modelling sources of variations in the same framework will increase the accuracy of the system in non-trivial viewing conditions.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Name Institute

Associated projects

Number Title Start Funding scheme
107959 Graphical Models pour l'authentification de visages 01.07.2005 Project funding (Div. I-III)

-