Rachid Deriche, three scientific lives in one

In 1979, freshly graduated from Télécom Paris, Rachid Deriche joined Inria in Rocquencourt to study digital processing and mathematical analysis of images.

"I was seduced by the fact of considering an image as a mathematical object, to be treated with models and equations.”

Computer science was then in its infancy. An image of 256 x 256 pixels was sometimes enough to saturate the memory of the few computers available! It was necessary to invent fast filtering techniques that required little computing power. The young researcher devoted himself to this task with some success. He develops a class of filters implemented very cleverly in a recursive way, in order to process images with a reduced and constant computational complexity, whatever the size of the filter.

The application of one of these filters (since known as the « Deriche filtre ») to the problem of contour extraction allows him to develop a fast and optimal algorithm in terms of robustness and localization, which is still in use and which has been taken up for real-time executions by different processors.

In 1988, Rachid Deriche joined the new Inria center in Sophia Antipolis, where he focused his work on computer vision and its applications in robotics. The problem was twofold: to observe their environment in 3D, the robots were equipped with two or more cameras, and these cameras filmed and therefore produced long series of images.

“With my colleagues in the Robotvis, project team, in particular Olivier Faugeras, we have cleared up several questions," says the researcher. “For example, how to calibrate and/or auto-calibrate a camera? How to reconstitute a 3D image from two or more 2D images? How to track and analyze motion from a sequence of images?”

Many of these tasks are natural for the human eye. But reproducing them with a computer is much more difficult. Rachid Deriche devotes himself to these tasks with passion, publishes extensively, and, with his colleagues, invents the key concept of the "fundamental matrix", which contains all the geometric information available and makes it easier to match two 2D images, to track objects and to self-calibrate.

In 1997, he also developed one of the very first algorithms for "mosaicking", i.e. assembling 2D views of the same scene from different angles to create its panoramic representation, a process used today in most of our smartphones. This innovation, as well as those linked to the development of 3D, gave birth in 1998 to the company Realviz (see box) and its special effects software for the digital image production industry.

In 2000, the researcher decided to devote himself to neuroimaging, as part of the Odyssey team. "The brain is fascinating, and despite remarkable advances, it is the unexplored continent of our time.

Taking this gamble means leaving a scientific community where he enjoys an international reputation, to join one where he is unknown, and where the acquisition systems and the physics of image formation are quite different.

But Rachid Deriche is thirsty for novelty and freedom." I tackled a set of new problems in neuroimaging to create innovative tools. They help to better explore and understand the brain via all of its structural and functional connectivity."

In 20 years, the researcher has made a name for himself in this field. Within the Athena team, which he created and directed until 2022, and in collaboration with several French and foreign laboratories and hospitals, he has made remarkable advances in cerebral connectomics.

 "The idea that each cerebral area is specialized in a specific task has been abandoned. In reality, several areas are networked for each task, and these networks are dynamic and evolve. Hence the importance of developing models, algorithms and software to reconstruct them, analyze them and, beyond that, study their dynamics."

To reconstruct and analyze the network of structural and functional connectivities, Rachid Deriche and his colleagues of the Athena project-team (Maureen Clerc, Théo Papadopoulo and Samuel Deslauriers-Gauthier) combine diffusion MRI, functional MRI, MEG and EEG. These techniques produce complex and heterogeneous signals at different spatial and temporal resolutions. The difficulty is to acquire them quickly, to process and analyze them correctly and to combine them in the best possible way to improve the solutions sought.

The work developed, in particular in diffusion MRI, has been used for both research and clinical purposes and notably transferred to the company Olea Médical, for their advanced MRI image post-processing tools. By revealing the network of connectivities to neurosurgeons, these techniques make it possible to increase the precision of surgical interventions for the treatment or complete removal of brain tumors, as demonstrated by Pr Maxime Descoteaux, his former doctoral student, and neurosurgeon David Fortin at the University and the University Hospital of Sherbrooke (Canada).

• Wikipédia page of Rachid Deriche
• Interview with Rachid Deriche on his different research arease (video), Inria, 18/11/2014.
• « From digital image processing and computer vision to neuro-imaging: A journey into geometric modeling », lecture by de Rachid Deriche, Canal-U, 18/12/2013.