TITLE: Facial Image Analysis from Video in Biometrics
AUTHORS: Jean-Luc.Dugelay, EURECOM, Sophia Antipolis, France.
ABSTRACT: The security field uses three different types of authentication: something you know, something you have, or something you are - a biometric. It ranges from fingerprints to retina, iris or facial characteristics. Biometrics is bound to become a part of our everyday life, playing a key role in enhancing security, residing in smart cards or passports and supporting personalized Web e-commerce services.
Face recognition (FR) is one of the most attractive biometrics for a broad range of applications. As it is the way people use to recognize each other, it is one of the most easily accepted biometrics. Furthermore, it is intuitive from the user's point of view
and doesn't need any contact.
2-D FR has been widely studied for decades and provides relevant performances. Nevertheless, such good results can only be obtained by applying tight constraints with respect to pose, illumination and expression of the face.
The major difficulty in FR is related to the existence of these three sources of variations. In this presentation, we will discuss on the possible contribution of video (i.e. dynamic features) for FR to improve performances of authentication while keeping existing advantages of face recognition from 2-D images. Combining with physiological parameters (i.e. related to the appearance) within a multimodal framework, behavioral parameters (i.e. related to head motion and facial mimics) would be helpful in the design of a person system more robust to pose, illumination and expression variations. This presentation will also include some points on very recent works on soft biometrics. It consists in extracting limited information from enrolled or even unknown persons such as gender, age, presence of eyeglasses, facial hair, color of eyes.
Finally, another major issue in biometrics is the security of the system, and not only its robustness. In the specific case of face recognition, one possible attack is the replay attack, i.e. the use of an ID picture from someone else.
F. Matta and J.-L. Dugelay
Tomofaces: eigenfaces extended to videos of speakers
ICASSP 2008, IEEE International Conference on Acoustics, Speech, and Signal Processing,
March 30 - April 4, 2008, Las Vegas, Nevada, USA.
F. Matta, U. Saeed, C. Mallauran and J.-L. Dugelay
Facial gender recognition using multiple sources of visual information
IEEE MMSP Cairns October 2008.