ODYSSÉE: Algorithmic and biological vision

Will progress in medical imaging make it possible to understand and model human or animal vision mechanisms in all their complexity ?

Modeling cerebral activity and its diffusion effects based on magnetic resonance images (MRI)

How does the brain see behind the eye ? As soon as attempts are made to mimic human vision, in robotics for example, numerous difficulties arise and it becomes clear that existing models only take into account an infinitesimal part of biological behavior. To enrich the understanding of human vision, the ODYSSEE team had the idea of exploiting the increasingly precise information supplied by imaging techniques: functional MRI (magnetic resonance imaging) that shows blood flows, diffusion MRI that displays structures through water diffusion, anatomic MRI that supplies a geometrical description, EEG (electroencephalography) and MEG (magnetoencephalography) that construct images based on the electrical activity of the brain. These techniques are making progress and are complementary. Some of them have an increasing spatial resolution, especially MRI, whereas others now provide a temporal precision on the order of the millisecond (EEG and MEG). Today, we can "see" the brain at work.

This is where ODYSSEE research begins: to go beyond case by case observations, exploit the detailed information contained in these images and renew vision modeling for a better understanding. One of the scientific challenges met by researchers is to integrate the different technologies in order to have the benefits of both spatial and temporal precisions in the same model. Using the models developed by ODYSSEE, it is now possible for example to extract RMI data from MEEG signals. ODYSSEE also widens the scope of its work to reach a more global understanding of vision by integrating neuron group or cortical column models.
Project ODYSSEE chose to associate the study of algorithmic vision with that of biological vision. The interaction of these two fields is fruitful for both. How does the human brain reconstitute a complete form when it only has a partial view of it ? How does it learn to recognize such and such a shape ? By studying the richness of biological vision, the team tries and models the problems of artificial vision (movement analysis, object recognition, shape representation and learning, etc.) using such mathematical tools as variational methods and partial differential equations.
The work of ODYSSEE results in the design of new algorithms and make it possible to devise new artificial vision tools by improving their robustness and reliability. This is of interest to many partners all over the world-universities, research centers, industry. ODYSSEE participates in different international projects on vision (Insight 2+ European project), image (Biomedical Image Analysis Canadian project) and neuroscience (Computation Tools for Brain Research with the University of Minnesota in the United States).

Practical information

Project launched in 2003, following the Robotvis team

Partners: INRIA, Ecole Nationale des Ponts et Chaussées (CERMICS), Ecole Normale Supérieure de Paris.
Research directions

• Variational methods and partial differential equations for vision
• Brain observation by functional imaging
• Modeling cortical activity.

The team
Seven research scientists (INRIA, ENPC, ENS)
Fifteen doctoral candidates and post-docs

Scientific Head
Olivier Faugeras