Here you will find :

    - My fields of interest
    - Some results and demonstrations from my PhD thesis
    - My PhD Thesis (ps.gz, PDF)
    - Some usefull publications available on the web
    - Other personal links to computer vision web pages

Back to CEREMADE front page

My advisor Laurent D. COHEN's home page

    - Multiscale methods
    - Signal and image matching
    - Stereovision
    - Image registration
    - Motion estimation
    - Shape deformation
    - Non linear inverse problems, non linear regression
 

    The purpose of my PhD dissertation

          It is to clarify the contribution of multiscale methods in motion estimation
    and registration algorithms and to improve them. My work might also help dealing
    with more general problems like non linear inverse problems and non linear regression.
 

    Abstract en Francais :

Dans ce travail nous abordons les problemes de l'estimation du mouvement apparent et du recalage de signaux et d'images sous hypothese de conservation des niveaux d'intensite. Dans une premiere partie nous introduisons une nouvelle hypothese de rigidite du mouvement et une pyramide multiechelle adaptee permettant d'ecrire une nouvelle equation de contrainte du mouvement valide aux echelles grossieres pour des mouvements d'amplitude importante. Nous en deduisons des conditions suffisantes de convergence d'un algorithme iteratif d'estimation/recalage au sens de la disparition du flot optique residuel et du recalage de l'image d'etude vers l'image cible, en dimension quelconque. Dans une seconde partie nous discutons la mise en oeuvre de l'algorithme et tentons l'approche par ensembles de niveaux en dimension 1 et 2. Ensuite nous obtenons des resultats satisfaisants en utilisant l'algorithme du gradient conjugue. Nous illustrons l'approche a l'aide de nombreux exemples synthetiques ou reels en dimension 1 (y compris sur paire stereo) et 2.

    Postscript version : these.ps.gz

    PDF version : these1.pdf (1st part) and these2.pdf (2nd part)
 

Presented in 1997 :

    - ICIP'97 (Eng., ps.gz) : A Multiresolution Algorithm for Signal and Image Registration
    - Gretsi'97 (Fr., ps.gz) : Un Algorithme Multiresolution de Recalage de Signaux et d'Images
    - ORASIS'97 (Fr., html) : Un Algorithme Multirésolution de Recalage de Signaux et d'Images

Searching for a "no a priori motion" model

                                            My subsequent work aimed at extending and generalizing these results without
    any motion model. I have then developped a multiscale eulerian motion estimation technique, based on a
    generalization of my first work. The mathematical study yielded a new non local motion estimator, and I
    give a proof of convergence of the method towards the real motion if it satisfies some rigidity hypothesis,
    together with the registrated image.

                                            As a first attempt to get rid of any a priori motion model, some numerical
    results using 1D level set boundaries are shown. For grey levels are not stable in most of the real
    examples, I then reconsidered the same motion model as an implicit one, which I successfully
    experimented on a wide variety of signals and images.  Although it is fully non local, the result should
    be compared to those obtained with the fluid registration process of Miller and Christensen, or
    Bro-Nielsen, or with Thirion's demons method, or with Trouvé's gradient algorithm.

Matching the same level sets : some demonstrations on boundary motion estimation in 1D

    - 1D signals
    - Grey level registration between two stereo images

Matching signals and images using an implicit motion model

    - 1D signals
    - Disparity estimation with stereoscopic Images
    - Images

English :

   - A. Trouvé : "An Approach of Pattern Recognition through Infinite Dimensional Group". A unified
        mathematical framework for many pattern recognition tasks with algorithms whose convergence are proved.
   - J-M Odobez, P. Bouthemy : "Robust multiresolution estimation of parametric motion of models
        applied to complex scenes". Many technical reports available at the IRISA Documentation Center
   - Gary Christensen, RD Rabbitt, MI Miller : "3D brain mapping using a deformable neuroanatomy".
   - M. Irani and S. Peleg : "Motion Analysis for Image Enhancement: Resolution, Occlusion, and Transparency".
   - Theodore Papadopoulo : a PhD thesis on "Motion analysis of 3D rigid curves from monocular
        image sequences".
   - Joao Costeira and Takeo Kanade : "Multi-body Factorization Method for Motion Analysis".
        Introduction of a new tool.
   - M. Bro-Nielsen : PhD thesis "Medical image registration and surgery simulation". On
        dynamical estimation-registration.
   - E.Mémin, P.Pérez, D.Machecourt : "Dense estimation and object-oriented segmentation
        of the optical flow with robust techniques".
   - N. Paragios and G. Tziritas : "Adaptive detection and localization of moving objects
        in image", among other numerous publications.
   - Z.Zhang, J.Feldmar, N.Ayache...: many technical reports on matching or registration at INRIA
   - C. Dimou and I. Pitas : "Parameter Estimation of Non-Translational Motion Fields".
   - Fadil Santosa : "A level-set approach for inverse problems involving obstacles". Self-speaking.
   - D. Bresch & J. Simon : "On the normal variations of a domain".
   - M. Covell : "Autocorrespondence : Feature-based Match Estimation and Image Metamorphosis" : motion and
         morphing.

French :

   - Zhang, Zhengyou : a state-of-the-art report on general matching techniques, among other publications.

A moving group (in France)

     - GDR-PRC ISIS (Research Group for Information, Signal, Images and ViSion)

Some links to general information on computer vision applications:

     - Machine vision
     - Fewer Factory Inspectors, More Electronic Eyes
     - Computer Vision in Postal Automation
     - Finding Images/Video in Large Archives
     - StereoVision : a tutorial