Name: Stéphane Derrode
S. Derrode received the telecommunication engineering degree from Télécom Lille in 1995, his PhD degree in signal and image processing from University of Rennes, in 1999, and his Habilitation à Diriger des Recherches from University of Marseille, France, in 2008. From 1999 to 2001, he worked as a research engineer at Télécom Bretagne, Brest (France). Since September 2001, he has been with the École Centrale Marseille where he is currently an Associate Professor in the Multidimensional Signal Processing Group, Institut Fresnel (CNRS UMR 6133), France.
During year 2005, he was a visiting researcher at the Cristal Laboratory / ENSI, Tunisia. He was a member of the Program Committee for the Taima’09 symposium, hold in Hammamet, Tunisia, in May 2009. His research interests include invariance and group theoretical image representation for pattern recognition and image indexing, and Markov models for image segmentation.
· W. Ketchantang, S. Derrode, L. Martin and S. Bourenanne, Pearson-based mixture model for color object tracking, Mach. Vis. and App., Vol. 19(5-6), October 2008, Special issue on video surveillance research in industry and academia.
· L. Martin, W. Ketchantang, S. Derrode, Method and device for selecting images in a sequence of iris images received in a stream, US Patent Application 2008/0075,335, STMicroelectronics SA and Université Paul Cézanne Aix-Marseille III. March 27, 2008.
Title of Project: A low-cost video-based iris recognition system
Among existing biometrics, iris-based recognition systems are recognized as one of the top most accurate personal identification systems. However, the acquisition of a workable iris image requires a strict cooperation of the user which can only be reached in controlled and fixed environments (e.g. high security building access). Trying to transfer this biometry in a mobile context for wide public applications (e.g. mobile phone unlocking) results in poor recognition performances due to degraded images (motion and focal blurs, partial occlusion due to blink…).
To overtake this limitation, we developed a complete biometric system based on a low-cost video acquisition of user’s iris. The selection of high quality workable iris images is done “on the fly” at the frame rate. To achieve a good level of accuracy, we implemented an algorithm to track users’ pupil (based on Kalman filtering) and defined a new spatio-temporal iris quality criterion incorporating eye relative velocity. For example, images of iris acquired during saccadic eye motions are rejected since they are more subject to blur than those acquired during fixations, especially when using a low-cost camera. The encoding, based on J. Daugman’s iris code, makes use of the Fourier-Mellin transform which is suited to take into account rotation of user’s head and dilatation of its pupil.
Performances of the different steps toward recognition, i.e. iris localization, image selection, iris signature extraction, are systematically analyzed by means of experiments on a video database. The high quality iris images extracted by the process are then used together to improve the accuracy and reliability of the recognition system.
Acknowledgment : The work has been conducted with W. Ketchantang (SAGEM Sécurité), L. Martin (ST MicroElectronics) and S. Bourennane (Institut Fresnel). Authors would like to express their gratitude to PACA region (France) and to ST Microelectronics for financial support.