Biometrics research activities
Biometrics has for objective to identify an individual or to authenticate him given different morphological, biological or behavioral characteristics. This process already used for different well known applications (Biometric USB key, logical and physical access control...) will be potentially the solution in the future to authenticate one client for an e-payement or others applications. The research activities held in this domain in the "E-payment & Biometrics" research unit are divided in three topics:
Definition of operational biometric systems: Many biometric systems have been developed (face, palm veins, keystroke dynamics, touch screen interaction, finger knuckle...) in this research unit. We use as low as possible data for the enrollment step and we try to use common and low cost sensors such as webcam, keyboards... We proposed a keystroke dynamics method using only 5 captures for the enrolment and providing an EER near 6%. In order to decrease recognition errors, we are also interested in multibiometrics. We defined new score fusion functions for multibiometrics providing a significant benefit compared to the state of the art based on genetic programming. Some recent works (Romain Giot's PhD thesis) focus on the update of the biometric reference of an individual in conjunction with B. Dorrizzi (Telecom SudParis).
Evaluation of biometric systems: Before using this technology in an operational context, it is necessary to evaluate (Mohamad El Abed's PhD thesis) and test it in real conditions. We have to take into account the performance of the biometric system but also its security and acceptability (social acceptance). We work with psychologists in this last approach. Some works are in progress in the definition of a generic quality metric of a biometric capture.
Protection of biometrics: We work on privacy issues by defining biohashing algorithms that consist in generating a biocode computed from a biometric reference that permits to revoke it if it has been stolen or after a certain period. We used this approach on fingerprints and are able to generate a biocode of size 384 bits while achieving a very performance from texture and minuatiae features (Rima Belguechi's PhD thesis). We defined also some secure frameworks to store biometric templates on smartcards using as for example Global Platform specifications in cooperation with the E-payment research activities in the unit.