Sites Inria

Version française

Inria Awards 2013

Jean-Michel Morel: 2013 lnria - French Académie des Sciences Grand Prize

Jean-Michel Morel

He used to want to do theory; he is now enthused by the applications of his work. He used to want to stay in academia; his algorithms are used by many industries and professional arenas. But Jean-Michel Morel is convinced that without mathematical formalisation, any discovery or theory remains incomplete. On the basis of this conviction, he uses mathematics for image and signal processing to explain the advances made and as the basis for theories. He does so successfully. He wants to share his algorithms freely, and to this end has set up a journal enabling anyone to experiment with other people's work.

Jean-Michel Michel opted for research in the mathematical field primarily because he didn't want to "concern myself with applications or have links with industry.  I preferred to be closer to theoretical work, which offered me more freedom!" Years later, the feedback from industry regarding his mathematical work, and the software and cameras that include his algorithms, shows that applications and industry have well and truly caught up with him! "It must be said that since I was doing my studies, both industry and its relationship with universities and research have changed a great deal," he acknowledges. "I find I now have great enthusiasm for the problems that manufacturers present me with. They send me their images and, using available computer technology, I am able to experiment with them immediately and start thinking about the solution straight away. In addition, their questions help me to design future research programmes."

Jean-Michel Morel works on image and signal processing at the Centre for Mathematic Studies and their Application (CMLA), part of the prestigious higher education and research institution, the Cachan École Normale Supérieure (ENS). His work has contributed greatly to the mathematical formalisation of digital image processing. "I am concentrating these days on the development of ever more effective algorithms to carry out very tangible tasks," he says. One of his areas of expertise is noise reduction, which improves the quality of digital images by removing the grain. "Algorithm-based image processing creates nothing that was not already in the image, but it improves focus, reduces chromatic aberrations, corrects distortions, etc. It means images are clearer and provide more information."

Such an application is of interest to the general public, which is generating more and more images with digital cameras. Image processing algorithms improve the quality of photographs taken at night or indoors, for example. Jean-Michel Morel's team has worked in this area with the company DxO, which designs and implements image enhancement algorithms for amateurs and professionals alike.

But his work is also much used in many professional contexts such as the interpretation of satellite images, medical imaging, etc. Jean-Michel Morel's team has worked for many years with the French National Centre for Space Studies, for which it has developed processing and analysis tools for aerial and satellite images. Used for the SPOT 5 and Pleiades programmes, these tools have been put to use in the detection of roads, buildings, etc. They are also used to produce digital models of areas of land based on stereo pairs from Pleiades satellites.

Another area of expertise for Jean-Michel Morel's team is image analysis. "What interests me in an application, such as face recognition, for example, which all digital cameras now have, is understanding why it works for faces but not for other things. The role of mathematics is to explain why it works. Science cannot restrict itself to merely inventing – the theory behind an experiment has to be established."
Jean-Michel Morel is also trying to "reform" the distribution of image processing algorithms both within the scientific community and outside it. He notes that the passage of an algorithm from its design phase to industrial application, its use in a software system or camera, is becoming faster. In addition, images abound, and are increasingly easier not only to take, but also to store and process, and so on. "All this progress has been haphazard, because everything is moving very quickly.  The web enables rapid and widespread distribution of the results obtained.  I wanted to establish a baseline for image processing algorithms that is verifiable online," he explains.
He has therefore started an internet journal (IPOL – image processing on line), in which authors make their algorithms directly available online in the form of pseudo-code, software and executable code. Anyone can consequently test the algorithms developed, on their own images, regardless of their machine or operating system. After images, he is hoping to broaden this initiative to sound processing, and then to various application sectors such as medicine, biology, etc.

Testimonial

Frédéric Guichard, co-founder of the company DxO, where he is the scientific director.
One of Jean-Michel Morel's PhD students.

"The algorithms developed by Jean-Michel Morel work extremely effectively in image processing, the field in which DxO, the company which I co-founded in 2003, specialises. His highly theoretical approach helps us to see the limits of the technology we offer and to anticipate developments in the market's requirements. The miniaturisation of cameras built into mobile telephones, for example, demands powerful embedded algorithms, which Jean-Michel Morel is helping us to develop.

I have worked with him continuously ever since he was my thesis supervisor. DxO is both a supplier and competitor to the R&D departments of the major camera manufacturers. Thanks to our collaboration with leading-edge research teams such as Jean-Michel Morel's, we are in a position to bring innovative products to the market every year. Engineers do not always have the technical answers to the problems they meet, and in formalising the scientific angle, Jean-Michel Morel helps them find solutions.

Keywords: Jean-Michel Morel Inria Awards

Top