Jean-Yves Tourneret

Professor at INP-ENSEEIHT Toulouse

Menu:

• Home
• Research
• Publications
• Teaching
• Professional Activities

Research activities

His research activities are centered around statistical signal and image processing, with a particular interest on Markov Chain Monte Carlo (MCMC) methods and on Detection and Classification.

Publications

List: here

MCMC Methods

Markov Chain Monte Carlo (MCMC) methods are simulation methods which can be used efficiently to solve problems inherent to Bayesian estimation and classification.  The original studies conducted by Jean-Yves Tourneret in the Signal and Communication group of IRIT are summarized below

• Changepoint detection and segmentation
  Hierarchical Bayesian algorithms combined with MCMC methods can be used efficiently for changepoint detection. These algorithms have been applied to the edge detection in  radar images, the joint segmentation of astronomical time series or the detection of changes in wind date. These works have been carried out in collaboration with Michel Doisy (university of Toulouse, France),  Nicolas Dobigeon (University of Toulouse), Marc Lavielle (university Paris) sud and Jeffrey Scargle (Ames Research Center, NASA, Moffett Field).
  
• Bayesian learning for classification and pattern recognition
  MCMC methods can be used to estimate the posterior distribution of a parameter vector conditionally to the different classes. The resulting estimated posteriors are then plugged into the Bayesian classification rule. Interesting results have been obtained for the classification of non-stationary signals, or more recently for the classification of altimetric signals. This work was carried out in collaboration with M. Davy (INRIA, Lille), and C. Doncarli (IRCYN, Nantes).
• Classification of digital modulations
  MCMC methods can be used to estimate the class-conditional distributions of linear and/or nonlinear digital modulations. The resulting Bayes classifier provides good performance when compared to other classifiers. This work was carried out in collaboration with Petar Djuric (university of Stony Brook, New-York), Thierry Robert (CNES, Toulouse) and Nathalie Thomas (University of Toulouse).  The PhD of Anchalee Puengnim was related to this  problem.
• Navigation and localization
  Navigation is receiving increasing interest in Toulouse. Sequential Monte Carlo techniques (Kalman filters, Unscented Kalman filters, particle filters, ...) are known to be interesting tools for estimating the postion of a fixed or moving vehicle. Our main activities within the group have been to develop efficient techniques for mitigating multipath effects and interferences. The PhDs of Jean-Rémi De Boer and Mariana Spangenberg have been carried out on this topic. This activity has been conducted with many collaborators including
  
• Vincent Calmettes from ISAE-Supaéro, Toulouse
• Anne-Christine Echer and Olivier Julien from ENAC, Toulouse
• Audrey Giremus from the university of Bordeaux
• Arnaud Doucet from the university of Bristish Columbia, Canada
  
• Josep Vidal from the universitat poytecnica de Catalunya (UPC), Barcelona, Spain
• Bertrand Lesot and Charles Dussurgey from Thales avionics, Valence
• Grégoire Duchateau from Thales Alenia Space, Toulouse
• Radar signal processing
  In the context of space time adaptive processing (STAP), Bayesian algorithms can be developped to mitigate clutter heterogeneity.  This work was conducted in collaboration with Olivier Besson and Stéphanie Bidon, both from ISAE-ENSICA, Toulouse.
• Hyperspectral Imagery
  Bayesian algorithms combined with MCMC methods can be used for the so-called linear unmixing problem which is classically used for hyperspectral images. The main advantage of these algorithms is to provide the full posterior of the abundances which can be for instance used to derive confidence intervals.  Extensions to semi-supervised algorithms which allow us to determine appropriate endmembers belonging to a known library are very promising. These works have been made in collaboration with
• Nicolas Dobigeon from INP-ENSEEIHT, Toulouse
• Chein-I Chang from the university of Maryland, Baltimore County, USA
  
• Eric Le Carpentier, Jérome Idier and Saïd Moussaoui from IRCCYN, Nantes
  
• Altimetry
  Bayesian algorithms combined with MCMC methods can be used efficiently to estimate the parameters of altimetric waveforms.  Using appropriate prior information allows to improve estimation performance when compared to usual methods such as the maximum likelihood method. This work was conducted in collaboration with Laiba Amarouche and Pierre Thibaut both from Collect Localization Satellite (CLS), Toulouse and with Corinne Mailhes from INP-ENSEEIHT, Toulouse.

Detection and Classification

Detection and classification have been important research topics in the group for a long time. Here are summarized some recent activities

• Change detection in optical and radar images
  This activity consists of studying multivariate distributions in order to estimate the correlations between different kinds of images. It has been conducted in collaboration with
• Florent Chatelain from the GIPSA-lab in Grenoble, France
  
• André Ferrari from Hyppolyte Fizeau laboratory in Nice, France
  
• Jordi Inglada from CNES in Toulouse, France
  
• Gérard Letac from the laboratory of Statistics and Probability (LSP) of Toulouse, France
  
• Philippe Réfrégier and Muriel Roche both form Institut Fresnel in Marseille, France
• Sophie Lambert-Lacroix from the university Joseph Fourier, Grenoble, France.
  

• Anomaly detection in Internet Traffic
  Multivariate gamma distributions are  interesting distributions which allow us to measure the dependence between random variables marginally distributed according to gamma distributions. The application of these distributions to the detection of anomalies in Internet traffic has been recently investigated. This work was conducted in collaboration with Patrice Abry and Pierre Borgnat, both from ENS Lyon, France and with Florent Chatelain from the GIPSA-lab in Grenoble, France

• Channel change and double-talk detection for echo cancellation
  Recent studies conducted in the group can be used to detect channel changes and double-talk for echo cancellation. This work was conducted in collaboration with Neil Bershad (Irvine university, CA, USA) and Jose Carlos Bermudez (Federal University of Santa Catarina, Brazil).
  

Publications

List: here

PhD Students

 Current

• Jérôme Séverini (2006-2009), on the classification of altimetric waveforms.  Co-advisor : Corinne MAILHES (INP-ENSEEIHT, Toulouse).
• Jean-Rémi De Boer (2006-2009), on GNSS-MEMS integrated navigation systems (suppported by DGA). Co-advisor : Vincent CALMETTES (ISAE-Supaéro, Toulouse).
  
• Olivier Eches (2007-2010), on MCMC methods for hyperspectral imagery. Co-advisor : Nicolas Dobigeon (INP-ENSEEIHT, Toulouse).
• Reza Shirvany (2008-2010), on change detection algorithms for heterogeneous data. Co-advisor : Marie Chabert (INP-ENSEEIHT, Toulouse).
• François Deudon (2008-2010), on wideband space time adaptive processing (supported by DGA). Co-advisors : Olivier BESSON (ISAE-Ensica, Toulouse) and Stéphanie BIDON  (ISAE-Ensica, Toulouse).
• Jean-Philippe Goy (2008-2010), on the use of parametric modelling for FMCW radars (supported by Rockwell Collins).  Co-advisor : Patrice Michel (TéSA laboratory, Toulouse).
• Cécile BAZOT (2009-2011), on Bayesian methods for gene analysis. Co-advisor : Nicolas Dobigeon (INP-ENSEEIHT, Toulouse).
• Chao LIN (2009-2011), on the detection of T-wave alternans in ECG signals (supported by TéSA laboratory). Co-advisor : Corinne MAILHES (INP-ENSEEIHT, Toulouse).
• Raoul PREVOST (2009-2011), on the separation of AIS signals (supported by DGA and CNES). Co-advisors : Martial Coulon (INP-ENSEEIHT, Toulouse) and David Bonacci (TeSA laboratory, Toulouse).
  

Past

• Mariana Spangenberg (2006-2009), on safe navigation for vehicles (supported by Thales Alenia Space). Co-advisors : Vincent CALMETTES (ISAE-Supaéro, Toulouse) and Olivier JULIEN (ENAC, Toulouse).
• Anchalee Puengnim (2005-2008), on the classification of digital modulations (supported by TeSA laboratory), currently working in TeSA laboratory. Co-advisor : Nathalie THOMAS (INP-ENSEEIHT).
  Phd Manusript :
• Stéphanie Bidon (2005-2008), on the estimation and detection in heterogeneous environment and the application to space time adaptive processing (STAP) (supported by DGA), currently working in ISAE-ENSICA. Co-advisor : Olivier BESSON (ISAE-Ensica, Toulouse). Phd Manusript :
• Florent Chatelain (2004-2007) on multivariate gamma based distributions for image processing applications, currently  assistant professor in the university of Grenoble.
  Phd Manusript :
• Nicolas Dobigeon (2004-2007), on hierarchical Bayesian models for multi-sensor processing,  currently  assistant professor in the university of Toulouse (INP-ENSEEIHT). Phd Manusript :
• Audrey Giremus (2002-2005), on the application of particulefiltering to GPS navigation supported by DGA), currrently assistant professor in the university of Bordeaux. Co-advisors : Vincent CALMETTES (ISAE-Supaéro) and Anne-Christine ESCHER.
  
• Patrice Michel (1997-2002), on the detection and estimation of modes in aircrafts with flexible structure (supported by Airbus, Toulouse, France), currently working in TeSA laboratory.
• Souad Meddeb (1996-2000), on the identification of Volterra and bilinear systems for communications.
  
• Martial Coulon (1996-1999), on the detection of parametric models in presence of additive and multiplicative noise, currently assistant professor in the university of Toulouse (INP-ENSEEIHT).
• Marie Chabert (1994-1997), on the time-scale detection of changes in synthetic aperture radar images, currently  assistant professor in the university of Toulouse (INP-ENSEEIHT).

© 2008 Jean-Yves Tourneret
Original design by Andreas Viklund