Nos partenaires

CNRS

Rechercher





Accueil du site > English > Research Topics > Topic 1 - Information Analysis and Synthesis > TCI Team

TCI Team

Traitement et Compréhension d’Images
Head : Hadj Batatia

 

The research activity of the TCI team focuses on Images, as spatial, temporal, and/or spectral representation of physical phenomena. The team is interested in models and methods for image processing and interpretation in three different domains : medical imaging, remote sensing, and the French sign language (LSF). The Image (the team’s main research topic) is addressed by formulating image processing problems such as :

  • reconstruction, restoration, super-resolution
  • segmentation, classification
  • registration, change detection, motion analysis, fusion

Throughout these problems, members for the team are interested in a set of shared challenges inherent to the image formation process and to the underlying physical phenomena, particularly :

  • data quality and variability : 1) noise, aretfacts, missing data ; 2) inter-region, inter-patient, inter-scene variability
  • multiple constraints : 1) spatial, temporal, spectral ; 2) regularity, singularity, sparsity
  • dimensionality curse : very large number of parameters, insufficient datasets
  • multi-modality : 1) coherence and complementarity ; 2) incompatibility in spatial, temporal and spectral scales

To handle these challenges, the team develops methodological tools for modeling and computing, particularly :

  • modeling : statistics, geometry, graphs, PDE, Bayesian methods
  • computing : optimization, sampling, parallel computing, regularization

The specific aspect of the TCI research lies in the unified approach adopted throughout the three domains. Precisely, a consistent scientific approach is considered for medical imaging, remote sensing and LSF, where models and methods are developed based on the underlying physical phenomena. Consequently, the team aims at developing shared methodological tools for the three domains. This specific aspect is illustrated through the following current research topics :

  • Bayesian methods and stochastic simulation methods, variational approaches and convex optimization for medical image processing and remote sensing
  • Parallel iterative methods for dynamic medical image filtering with strongly non-linear EDP models for convection-diffusion
  • Information geometry for preconditioning optimization problems applied to parameter estimation and medical image segmentation
  • Optical and radar (satellite, embedded sensors) image processing for automated guidance of groups of vehicles (cars, drones...)
  • Elaboration of domain specific sign-lexicon and gesture recognition for LSF

Medical imaging

In medical imaging, the team is interested in anatomic (CT, MRI, HFUS, OCT, Confocal microscopy) and functional (PET, fMRI, MrsI, EEG) images. Two application domains are the most frequently addressed : oncology and brain activity. For oncology, the team currently studies subjects such as tissue heterogeneity for tumor characterization, organs and tumors segmentation and evolution, therapeutic assessment and follow-up, tumor response and relapse prediction, and physiological motion compensation. For brain activity, the team conducts research projects on epilepsy and prediction of awakening in traumatized comatose. The team is particularly interested in dynamic images (3D + t) and multi-modal functional images.

Remote sensing

In remote sensing, the team is interested in multi-modality for change detection, unmixing, and guidance of groups of vehicles. Image modalities currently under study include radar (SAR, GPR), hyperspectral, optics and infrared.

French sign language

The French sign language makes a large research field that covers different and complementary domains, such as illustrated in the following figure :

 

 

In this domain, TCI is interested in the analysis and interpretation of communication gestures and attitudes made by persons observed with a vision system. The aim is to detect and track body components (hands, face, chest), to reconstruct the persons’ 3D posture, to analyze expressions, to finally make the linguistic interpretation of this information by recovering the semantics. To account for the variability of the human body and the complexity of the sign language, prior knowledge and constraints are incorporated throughout the process. Accordingly, the team’s research activity concentrates on dynamic models for gestures and face expression, and models of the sign language ; while exploiting these models in interpretation methods. The team works also on designing and analyzing sign lexicons for specific domains, such as mathematics.

Cette rubrique ne contient aucun article.