Project Team Moise

Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry
Partnerships and Cooperations
PDF e-pub XML

Section: New Results

Image processing

Segmentation and assimilation of medical images

Participant : Nicolas Papadakis.

In collaboration with the Inria team MC2 of the Bordeaux-Sud-Ouest center, we investigate the application of image assimilation to medical issues. The objective is here to use MRI images in order to monitor EDP models dealing with tumor growth in lungs or brains. Using such images, we would like to define a patient specific process allowing to calibrate the numerical model with respect to the observed tumor. First works based on convex relaxation of the binary segmentation problem [81] have been realized in this direction by proposing a 3D segmentation method dedicated to glioblastomas from a set of MRI brain images. The obtained automatic segmentation results are very close to specialist manual segmentations (errors of 5%) and will be used as pseudo-observations for an assimilation system based on the numerical model describing the tumor growth. The final issue will be to define an observation operator linking images with the model in order to realize a direct assimilation.

Optimal transport

Participants : Maƫlle Nodet, Nicolas Papadakis, Arthur Vidard.

A new activity on optimal transport has been started in collaboration with the EDP and MGMI teams of the Laboratoire Jean Kuntzmann, Grenoble and the MAP5 Laboratory, Paris, through a project funded by the ANR white program. The purpose is to define metric beween images involving the so-called Wasserstein distance. Such metric would be of particular interest in order to introduce pertinent observation operators for assimilating image data 6.4 . Other applications including image morphing and histogram equalization are also studied for image processing purposes. First results has been obtained by adding spatial regularity to the transport map computed by the Benamou-Brenier algorithm [66]

Computer vision

Participant : Nicolas Papadakis.

In collaboration with Vicent Caselles (Pompeu Fabra University, Barcelona, Spain), different image processing works have been finalized: 3D reconstruction and novel view synthesis for soccer replays [13] , stereo inpainting for 3D movie post-production [29] . Other works dedicated to object tracking in image sequences have also been proposed with Aurelie Bugeau (LABRI, Bordeaux) [14] . A main attention is now given to the problem of histogram equalization of different images [15] . Our aim is now to include spatial information on color repartition during the histogram transfer for inpainting and shadow removal purposes. A journal paper dealing with this issue has been recently accepted for publication.