Postdoc Positions in Image Analysis at CEREMADE, PARIS, FRANCE
Postdoc in Laurent COHEN s group, starting September 2009 or early 2010, for 1 or 2 years on
(This page available at http://www.ceremade.dauphine.fr/~cohen/MVA/postdocMesange.html)
The problem to be solved is the extraction of coronary networks and characterization of their motions in a
rotational X-ray projection sequence. Rotational X-ray imaging devices consist of a rotation (over a wide angle)
of a source-detector system and provide a new way to explore the possibly existing artery pathologies enhanced
after injection of a dye product. Such imaging techniques should replace at least partially the examinations
performed from a few static, 2-D angiograms acquired from several static points of view by offering higher
accuracy for lesion assessment and improved planning for endovascular interventions. However, this problem
is a true challenge and departs from many computer vision applications. By principle, all the structures crossed
by the X-rays (bones, soft tissues) are cone-beam projected onto the 2D detector. Due to the rotation of the
source-detector system, all these objects are moving and their features (including the most basic one, the
intensity, e.g the attenuation) are varying over time and space. The object of interest, the coronary tree, is
patient dependent (changes in its branching structure, complex-shaped vessels), has low non-stationary contrast
and crossings, superimpositions in the image sequence planes represent additional difficulties to deal with. The
coronary motions, determined by the myocardium contraction, can not be reduced to rigid transformations:
strong deformations occur with slow and fast phases and movement inversions. In other words, and to
summarize the global picture, the proposed project has to face a multi-object, multi-motion, highly nonstationary
and low signal to noise problem in a (semi) real-time context. Aimed at segmentation, motion separation and modelling, they
include minimal path and fast-marching algorithms, topological-gradient based methods but also perceptual
grouping, string-matching, thin-plate spline techniques going through local to global approaches, dense motion
field estimation, feature-based descriptions (lines for example). The overall goal of the project is thus to
provide quantitative, robust and reliable information on object motions that would potentially feed
reconstruction methods. A specific, already relevant aim in terms of clinical added-value and industry transfer,
is to get a breakthrough for real-time catheter tracking in interventional setting.
Our part of this project will be in continuation of extensive work on minimal paths developed at CEREMADE these last years,
presented in particular in our recent publications at ECCV08, SSVM09, MMBIA09, CVPR09, ISBI09 and
available at Laurent Cohen s list of publications (http://www.ceremade.dauphine.fr/~cohen).
--------------------------------------------------------------------------------------------------------------------------------------
send me (cohen@ceremade.dauphine.fr) a CV, references, and a letter and we will discuss about more details.
DEADLINE: October 31st, 2009
CEREMADE is a large lab in Applied Mathematics, located in Paris, France,
with a team in Image Analysis working mainly on Variational methods and Partial Differential Equations for Image Analysis,
in particular on Minimal Paths, Fast Marching methods.
In order to know what's done at CEREMADE in Image analysis, you can
have a look at the list of publications:
http://www.ceremade.dauphine.fr/~cohen/publithematic.html
Location: University Paris-Dauphine, Paris, France. Metro and Bus Porte Dauphine, RER C Foch.