Awarded LABEX Postdoc Grant (information theory for surveillance).



PIs Petros ELIA and Francois Bremond (EURECOM/INRIA)
Postdoctoral scholar Dr. Antitza Dantcheva (see
Research Axes Big Data, eHealth


The project is motivated by the challenge of efficiently processing big data from facial images and facial biometrics. Given the massiveness of this data (currently running in the Exabytes), existing processing efforts fail in terms of reliability and speed. We believe that any progress in this area must come by understanding the true information content of such images, in essence understanding how unique and distinctive faces are. Towards this we propose a never-before-attempted exposition that combines our expertise in computer vision and information theory, towards understanding the entropy of faces, using custom-made `descriptive-complexity’ measures that we will design, that will be based on how long of a code is needed to describe different faces. The longer the code, the more unique these faces are. An important outcome of this approach will come from our ability to understand the true limits of processing big facial data which, in layman’s terms, could allow us to differentiate if suboptimal processing performance is due to the use of suboptimal algorithms, or if it is due to the fact that the class of facial images inherently does not have enough uniqueness /distinctiveness.

Special emphasis will be placed on applications that can benefit from this hybrid approach. One such application naturally relates to advancing our ability to efficiently process data in security-related surveillance networks. Our most promising application though is in the area of eHealth where we intend to apply our approach to a class of individuals that cause a particular challenge to facial recognition; the class of elderly people. Understanding the uniqueness / distinctiveness properties of elderly faces, will be of particular importance because often times, such elderly patients can indeed be assisted by such facial recognition, just as in the case of Alzheimer’s patients. Another application, which again will first focus on Alzheimer’s patients, will be to – using our proposed IT based approach – understand how distinctively different faces convey emotions. Understanding the limits of this particularly hard computer vision task of emotion recognition in elderly patients, will assist us in our existing efforts in the rehabilitation of such patients during, for example, music therapy sessions where automatically detecting emotional states offers cues for better treatment of such patients.