Aucun titre de diapositive

1
Overview of the talk
1. How everything begun or why am I here ?
2. Scientific organization at University Paris
Descartes focus on image processing research
3. Overview of image processing research projects
and collaborations at University Paris
Descartes 4. What next ? Scientific
Collaborations ? Student Exchanges ?
2
1. How everything begun ?
Georges Stamon
Volkan Atalay
Florence Cloppet
Nicolas Loménie
Nese Yalabik
Student exchange for short term period (one
month, 4 students, Ankara-gtParis)
Guray Erus, PhD student (Parislt-gtAnkara)
3
2. Scientific organization at University Paris
Descartes focus on image processing research
University Paris Descartes
http//www.parisdescartes.fr/

• 10 Teaching and Research Departments (UFR)
• Biomedicale
• Droit (Law)
• IUT (Technology)
• Maths Info (Applied Math. Computer Sciences)
• Médecine
• Odontologie
• Pharmacie
• Psychologie
• Sciences Humaines et Sociales
• STAPS (Sport)

4
UFR de Mathématiques et dInformatique Dpt. Of
Mathematics and Computer Sciences
http//www.math-info.univ-paris5.fr/

• 1500 students
• 60 Professors
• Two research laboratories
• MAP5 for applied mathematics
• CRIP5 for computer sciences

Within each of them, image processing and
computer vision issues in both teaching and
research
5
http//www.math-info.univ-paris5.fr/map5/
Christine GRAFFIGNE
Frédéric RICHARD
UFR de Mathématiques et dInformatique Dpt.
Of Mathematics and Computer Sciences
Site http//sip-crip5.org
http//www.math-info.univ-paris5.fr/crip5/
Nicolas LOMENIE F.X. JOLLOIS
Florence CLOPPET-OLIVA
Georges STAMON
NicoleVINCENT
15 members PhD., Ass. Professors, Professors.
6
Head

• Prof. Georges Stamon
• Prof. Nicole Vincent

Contacts Prenom.Nom_at_math-info.univ-paris5.fr
Assistant Professors

• Dr. Florence Cloppet
• Dr. Nicolas Lomenie
• Guray Erus

7
Thematical orientation Visual Perception with
AI elements (semantic filtering, intelligent
control, interpretation ...)
3. Overview and references of image processing
research projects and collaboration at University
Paris Descartes
Thematical orientation Mathematic modeling for
image analysis (statistic, Markov fields, EDP
...)
8

• Results
• More than 35 thesis defended
• More than 100 papers ( including PAMI, PRL)
• Monthly seminar
• Collaborations with
• IGN (National Geographic Institute - Paris)
• CNES (National Space Agency - Toulouse)
• Institut Pasteur (Paris)
• A2iA
• EADS
• SAGEM
• France Télécom

University of Montréal Canada Ecole des
Technologies Supérieure (ETS-Montréal)
University of Concordia (Montréal) University
of Ankara METU -Turquie Almaden Research Center
(IBM) - Californie
9

• 3 axes of knowledge
• Bio-imaging Institut Pasteur, Institut Pasteur
  Korea, Paris Descartes
• Image and Document analysis SAGEM, THALES,
  A2IA
• 3D and video analysis France Telecom, MKL
  Systems, Institut Pasteur

10

• Bio-imaging Institut Pasteur, Institut Pasteur
  Korea, Paris Descartes -gt Estelle Glory, Florence
  Cloppet, Sylvain Berlemont

Estelle Glory
Now at Carnegie Mellon University
The Center for Bioimage Informatics
Reference V. Meas-Yedid, E. Glory, E. Morelon,
Ch. Pinset, G. Stamon, J.C. Olivo-Marin,
Automatic color space selection for biological
image segmentation, Proceedings of ICPR, vol 3,
514-517, 2004.
11
Acquisition System inverted, motorized
microscope with computer control

• Inverted microscope
• objectifs x10, x20
• motorized stage (x,y)
• motorized z (focus)
• fluorescence possible adaptation

• Computer software
• control of plate moving
• autofocus
• control of camera acquisition
• image storage

• Digital camera
• color
• cooled

12
Image processing segmentation of nuclei
1. Thresholding

• magenta-colored nuclei
• ? green component
• grey level threshold
• size selection

Automatic threshold (iterative)
Isolated nuclei
Original image
Green component of rgb
Adjacent nuclei
13
Image processing segmentation of nuclei
2. Watershed

• Seeds creation
• - ultimate erosion
• - 2 dilations

Adjacent nuclei
Segmentation merging
seeds
watershed
14
Image processing segmentation of nuclei
3. Results
Testing of several conditions
B condition
C condition
A condition
D condition
15
Florence Cloppet
Reference F. Cloppet, J.M. Oliva, G. Stamon,
Angular Bisector Network, a Simplified
Generalized Voronoï Diagram Application to
Processing Complex Intersections in Biomedical
Images, PAMI-IEEE, vol 22 n1, janvier 2000,
p120-128.
16

• In collaboration with
• lUnité de Biologie Cellulaire du Parasitisme (E.
  Labruyère)
• LUnité dAnalyse dImages Quantitative
• (C. Zimmer)

17
Characterization of deformations

• Methodology

18
Characterization of deformations

• Localisation of deformations

Extremal Point (PE)
Branching Point (PB)
19
Characterization of deformations Outgrowth or
Retraction ?

• Info statique nest pas suffisante
• Utilisation de linfo dynamique entre T-1, T

Ir
Ig
PB
20
Ir
Ig
Ig
PB
Ig
Ir
PB
Ir
Ig
Ir
Outgrowth if  d(Ir, Ig) gt ? and
d(PB,Ir)-d(PB,Ig) gt0
No Evolution if  d(Ir, Ig) ? ?
Retraction if  d(Ir, Ig) gt ? and
d(PB,Ir)-d(PB,Ig) lt0
PE
PB
21
PE
PB
22
Characterization of deformations results
23

• Bio-imaging Institut Pasteur, Institut Pasteur
  Korea, Paris Descartes -gt Estelle Glory, Florence
  Cloppet, Sylvain Berlemont

What to keep in mind ? - classical issues of
image processing research set up in an
operational environment like start-up,
hospital - not so bad because the main issue in
bio-imaging is the acquition of a good quality
image database - strong link with Institut
Pasteur, and hospital with short term biological
issues
24
Zipf Law

• Image and Document analysis SAGEM, THALES,
  A2IA -gt Nicole Vincent, Guray Erus, Fares
  Menasri, Rabie Hachemi, Rudolf Pareti

• Nicole Vincent

• Reference
• 1. N. Vincent, A. Séropian, G. Stamon - Synthesis
  for handwriting analysis, Pattern Recognition
  Letters, Elsevier, N26, fév. 2005, pp 267-275
• 2. Y. Caron, P. Makris, N. Vincent. Zipf law
  models for image analysis, Fractals in
  Engineering V, 22-24 juin 2005, Tours, France.

25
Log (frequency)
- Empirical law - First applied to text analysis
more than 50 years ago
26
We need a specific alphabet of 2D patterns
(application driven) and then applied this
framework to - image coding - optimal
observation scale - image indexation - ROI
extraction ...
slope of the right part of Zipf plot
slope of the left part of Zipf plot
27
Detection and Recognition of cartographical
objects in VHR Satellite Images

• Image and Document analysis SAGEM, THALES,
  A2IA -gt Nicole Vincent, Guray Erus, Fares
  Menasri, Rabie Hachemi, Rudolf Pareti

• Guray Erus

Reference Extraction of cartographic objects in
high resolution satellite images for object model
generationGuray Erus Nicolas Lomenie 2006 - 4th
Workshop on pattern Recognition in Remote Sensing
in conjunction with ICPR2006 - Hong-Kong
28
Context
Objectives
Image samples for each class (100x100 pixels)
VHR Satellite Image (24000x24000 pixels)
29
Radiometric clues and rule-based model not
generic, not usable because of statistical
learning step....
Result of learning
Labelling to learn
Rule 1 IF two big water regions surround a
little region THEN the little one is a bridge
30
What about geometric clues and structural model
generic, usable ?
A few samples labelled by expert
Primitive Geometric Decomposition
31
Structural model learning via Graph Matching
techniques and a few examples
32
On going work about This structural model How
to use it ? How to edit it ? How to improve it
via user interaction ? How to swith to another
modelling framework as fuzzy linguistic rules in
order to enhance interactivity ?
33
And much more important but difficult how to
detect in a 24000x24000 VHR image
Obtenir des points d'ancrage ?
() En coopération avec Image Processing and
Pattern Recognition Laboratory, Dep. of Computer
Engineering, METU, Ankara, Turquie)
34

• Image and Document analysis SAGEM, THALES,
  A2IA -gt Nicole Vincent, Guray Erus, Fares
  Menasri, Rabie Hachemi, Rudolf Pareti

What to keep in mind ? - more original issues
of image processing research set up within an
institutional environment like University,
CNES - much more long-term research (several PhD
thesis)
35

• 3D and video analysis France Telecom, MKL
  Systems, Institut Pasteur

Auguste Genovesio
Korea
Now at Image Mining Group
http//www.ip-korea.org/
Reference Genovesio A, Liedl T, Emiliani V,
Parak WJ, Coppey-Moisan M, Olivo-Marin JC.
Multiple particle tracking in 3-Dt microscopy
method and application to the tracking of
endocytosed quantum dots. IEEE Trans Image
Process. 2006 May15(5)1062-70.
36
General methodology

• 2 steps

pairing
detection
t 1
t 1
t 1
t 2
t 3
t 4
t 5
t 6
Image sequence
Sequences of coordinates
Tracks
37
General methodology

• Pairing
• 2 steps prediction (IMM) then association

38
MKL Systems
Reference A. Auclair, L. Cohen, N. Vincent, A
robust approach for 3D model reconstruction from
a video sequence of cars,  In Proceedings of the
15th Scandinavian Conference on Image Analysis,
SCIA 2007, Aalborg, Denmark, June 14-17, 2007,
Lecture Notes in Computer Science, Springer,
Berlin, 2007 (to appear)
39
(No Transcript)