Image processing
Scientific referents
Coordinator : Jean-Michel Morel
Scientific leaders :
- Miguel Colom
- Agnès Desolneux
- Gabriele Facciolo
- Rafael Grompone
- Enric Meinhardt Llopis
- Jean-Michel Morel
- Alain Trouvé
Key figures
10 post-doc ; 14 phd students ; 3 internship students
70 phd theses defended in 2 decades
Presentation of the thematic area
The team led by Gabriele Facciolo and Jean-Michel Morel works on
-
Mathematical theory and algorithms for image and video processing,
-
Camera and satellite design,
-
Imagery,
-
Statistical detection theory applied to images and videos,
-
Variational restoration methods and applications of deep learning to computer vision.
The team led by Agnès Desolneux develops stochastic models for images, mainly for texture analysis and synthesis, and works on random field geometry.
The team led by Alain Trouvé develops sophisticated applications of geometric measure theory and Riemannian geometry for image comparison and shape modeling, with many applications to computational anatomy and molecular biology.
Key words
Image processing; Video processing; Image analysis; Video analysis; A contrario detection; Image authentication; Deep learning; Satellite imagery; Computational anatomy; Texture modeling; Computer graphics; Stochastic image models.
Key facts
- 2021 : Best student paper award for Nguyen Ngoc Long at the 2021 Workshop EarthVision (article co-authored with Jérémy Anger, Axel Davy, Pablo Arias and Gabriele Facciolo), "Self-supervised multi-image super-resolution for push-frame satellite images"
- 2021 : Best paper award at the International Workshop for Biometrics and Forensics (IEEE-IWBF 2021) for Marina Gardella, Miguel Colom, Jean-Michel Morel and Pablo Musé
- 2021 : Thesis award Prix de la Chancellerie 2020 from the parisian universities for Jean Feydy "Analyse des données géométriques, au-d elà des convolutions", director Alain Trouvé
- 2020 :Thesis award from the Association française de reconnaissance et d'interprétation des formes (AFRIP) for Jean Feydy "Analyse des données géométriques, au-d elà des convolutions", director Alain Trouvé
Structuring projects
IPOL (Image Processing on Line) is a novel journal that promotes reproducible research in image processing. Each published algorithm must include an online demo that can be run on any image uploaded by the user. In a decade, this journal founded in 2011 has published about 200 algorithms with a mathematical description, a computer code, an online demo, and a public experimental archive containing over 350,000 online experiments performed by users from over 2,000 different academic or industrial centers on five continents.
Applications
The Image Processing team does not only develop theories, but also accompanies their development and their long-term applications in industry and society in general. It collaborates in particular with :
- the french spatial agency (CNES) on the design of Earth observation satellites;
- The major American satellite manufacturer Planet on the image processing chain of its constellations;
- Laboratories DxO (world leader in image processing) on image denoising;
- Huawei on the next generation of mobile phones;
- The Agence France Presse and the french scientific police on detection of forgery images and videos;
- Kayrros on the automatic exploitation of earth observation satellites to detect oil and gas activities and sources of methane and CO2 emissions;
- Tarkett (a major carpet manufacturer) to design natural-looking textures;
- Thalès on the recalibration of satellite aerial images for the guidance of flying objects;
- SNCF on the "Detection of risky behaviors in railway stations by non-biometric video analysis"
The overall average of research funds obtained by the image processing team through competitive calls, grants and industrial contracts amounts to 1,100,000 euros per year.