La lecture en ligne est gratuite
Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres
Télécharger Lire

presented to obtain the

De
80 pages
Niveau: Supérieur

  • dissertation


Dissertation presented to obtain the Habilitation à Diriger des Re her hes Université Paul Sabatier Toulouse 3 Mention: Applied Mathemati s by Didier Auroux Fast algor i thms for image pro essing and data assimil a t i o n Defended on 26 Novemb er 2008 After reviews by: G. Aubert, Professor Université de Ni e Sophia-Antip olis P. Rou hon, Professor Mines ParisTe h F. Santosa, Professor University of Minnesota Commitee memb ers: G. Aubert, Professor Université de Ni e Sophia-Antip olis (Reviewer) J. Blum, Professor Université de Ni e Sophia-Antip olis (Examiner) L. Cohen, Resear h dire tor CNRS & Université Paris Dauphine (Examiner) P. Degond, Resear h dire tor CNRS & Université Paul Sabatier (Examiner) M. Masmoudi, Professor Université Paul Sabatier, Toulouse (Advisor) J.-P. Puel, Professor Université de Versailles Saint-Quentin (Examiner) P. Rou hon, Professor Mines ParisTe h (Reviewer)

  • exp eriments

  • mo dels

  • ation problem

  • top ologi al

  • pro essing

  • exp erimental approa

  • intro du tion

  • tion terms

  • numeri al


Voir plus Voir moins

Vous aimerez aussi

CNRS
D
G.
i
.
ss
ol
er
ersit?
t
sa
ati
R
o
Univ
n
P
p
Mines
r
P
e
C
se
de
n
,
te
L.
d
.
t
(Examiner)
o
s
o
tin
bt
l
a
n
in
h
t
ersit
he
ee
H
,
a
ol
b
J.
i
de
li
E
t
h
a
Da
ti
,
on
P
?
,
Di
oulouse
ri
Puel
g
es
er
R
de
h
s
s
Re
ou

Professor
her
T

S
Univ
Professor
ersit
of
?
mmi
P
b
a
uber
u
Univ
l
Sophia-An
Sabatier
s
T
r
oulouse
lu
3
Univ
Men
Sophia-An
tion
s
:
mi
Applied
,
Mathe
CNRS
mati
P

(Examiner)
b
gon
y
h
D
Univ
i
Sa
di
Masmoud
e
Univ
r
Sabatier,
A
A
ur
J.-P
ou
Professor
x
V
F
t-
a
P
s
,
t
arisT
a
eviewer)
l
i
g
P
o
R
r

i
,
t
Mines
h
aris
ms
ec
for
F.
image
anto
pro
,

Univ
ng
y
and
Minnesota
data
o
assimil
t
a
mem
t
ers:
i
A
o
t
n
Professor
De
ersit?
fended

on
tip
2
i
6
(
No
eviewe
v
)
em
B
b
m
er
Professor
2
ersit?
00

8
tip
A
i
f
(
t
xa
e
ner)
r
Cohen
re

v

ie
&
w
ersit?
s
aris
b
uphine
y
P
:
De
G
d
.

A

u
&
b
ersit?
e
aul
r
batier
t
M.
,
i
P
Professor
r
ersit?
o
aul
fe
T
ssor
(
U
dvi
n
or)
i
.
v
,
ersi
Univ
t
de
?
ersaill
de
Sain
Ni
Quen

(Examiner)
So
.
p

hi
Professor
a
P
-
ec
An
(R
tip
o.
C
.
o
.
n
.
t
.
en
.
t
T
s
.
1
.
I
to
n
.
t
.
r
.
o
.
d
.
u
.

.
ion
.
7
.
2
.
I
.
m
o
a
.
g
y
e
.
pro
.

.
ssing
.
11
.
2.
.
1
.
I
Extension
n
.
t
.
r
.
o

d
.
u
.

.
.
.
.
.
.
restoration
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
n
.
.
.
[26]
.
27
.
.
.
2.8.2
.
.
.
.
.
19
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
cation
.
.
.
.
.
.
.
21
.
problem
11
.
2.2
and
T
.
op
.
olo
.

.
al
tatio
as
.
y
.

.
analysis
23
.
.
.
.
.
.
.
expansion
.
.
.
.
.
24
.
.
.
.
.
.
.
.
.
up
.
.
.
.
.
Discrete
.
.
.
.
.
26
.
gradien
.
.
.
.
.
t
12
.
2.2.1
.
Pr
.
esen
.
tation
.
of
.
t
hing
he
.
metho
.
d
.
.
.
.
.
.
Remarks
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
19
.

.
.
.
.
.
.
.
.
.
.
.
2.5
.
0,
.
.
12
.
2.2.2
.
Main
.
result
.
.
.
.
.
.
In
.
the
.
.
.
.
.
.
.
2.5.2
.

.
.
.
.
.
.
.
2.5.3
.
ervised
.
.
.
.
.
.
.
2.6
.
[12,
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
F
.
segmen
.
.
12
.
2.
.
3
.
Inpain
2.6.2
ting
er
[
.
9,
.

.
.
.
.
.
.
.
.
Algorithm
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.7
.
sp
.

.
.
.
.
.
.
.
.
.
26
.
transform
.
.
.
.
.
.
.
.
.
.
.
Preconditi
.
d
.
.
.
.
.
.
13
.
2.3.1
27
Crac
ological
k
minimal
lo
.

.
zation
.
pro
.
blem
Minimal
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
ast
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
.
.
.
.
.
.
.
2.4.4
14
.
2.3.2
.

.
hlet
.
and
.
Neumann
.
form
.
ulations
.
for
.
the
.
inpain
.
ting
.
problem
.
14
.
2.3.3
.
Asymptotic
.
e
2.4.5
xpansion
to
.
images
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
19
.
Classi
.
[1
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
15
.
2.3.4
.
Algorithm
.
.
.
.
.
.
.
.
2.5.1
.
tro
.
to
.

.
.
.
.
.
.
.
.
.
.
.
.
.
21
.
Restoration
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
.
Extension
.
unsup
.

.
.
.
.
.
.
16
.
2.3.5
.
Remarks
.
.
22
.
Segmen
.
n
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.6.1
.
rom
.
to
.
tation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
17
P
2.4
w
Restoration
series
[16,
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.6.3
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
.
Complexit
.
and
17
eeding
2.4.1
[13,
V
.
ariational
.
form
.
ulation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.7.1
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.7.2
.
o
.
e
17

2.4.2
t
T
.
op
.
ological
.
gradien
.
t
.
.
.
.
.
.
.
.
2.8
.
op
.
gradien
.
and
.
paths
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.8.1
.
paths
.
.
.
.
.
.
.
.
.
.
.
.
18
.
2.4.3
.
Algorithm
.
.
.
.
.
.
.
.
.
.
28
.
F
.
marc
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
28
.
..
4
.
C
In
O
.
N
.
T
.
ENT
Conclusion
S
.
2.8.3
.
Coupled
.
algorithm
.
.
w
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
.
.
.
.
.
.
.
4.2
.
tness
.
.
.
.
.
.
.
.
.
.
.
.
29
.
2.9
.
Conclusions
.
and
.
p
.
ersp
.
ectiv
.
es
.
.
.
.
47
.
.
.
t
.
.
.
linearized
.
.
.
.
.
.
.
.
.
lation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
optimizatio
.
.
.
.
30
57
3
.

4.3.1
k
.
and
4.3.2
fort
.
h
.
n
.
udging
.
31
.
3.1
.
In
.
tro
.

ort
.
.
.
.
.
In
.
.
.
.
.
.
.
and
.
.
.
.
.
.
.
for
.
.
.
.
.
3.5.2
.
.
.
.
.
.
.
ergence
.
.
.
exp
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
.
tro
.
.
.
.
.
.
.
.
31
algorithm
3.2
.

.
k
4.2.1
and
.
F
.
orth
54
Nudging
.
(BFN
.
)
.
algorithm
55
[8,
.
11,
.

.
.
Muti-grid
.
.
.
.
.
y
.
.
.
.
.
.
.
erimen
.
.
.
.
33
.
3.2.1
.
F
.
orw
.
ard
.
n
data
udging
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Viscous
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
linear
.
.
.
.
.
.
.
.
.
.
.
.
.
Burgers
.
.
.
.
.
.
33
.
3.2.2
.

.
kw
3.5
ard
ers
n
.
udging
.
.
.
.
.
.
.
.
.
.
Observ
.
shallo
.
mo
.
.
.
.
.
.
.
.
.
v
.
terms
.
.
.
.
.
.
.
.
.
.
.
Con
.
on
.
system
.
.
.
3.5.4
.
ts
.
.
34
.
3.2.3
.
BFN
.
algorithm
.
.
51
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
data
.
4.1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
of
.
.
.
.
.
.
.
.
.
.
34
.
3.2.4
t

.
of
.
the
.
n
.
udging
.
matrices
Cost
and
.
in
.
terpretatio
.
n
.
.
.
.
.
.
.
.
Regularization
.
.
.
.
.
.
35
.
3.3
.

.
exp
55
erimen
h
ts
.
[11,
.
14,
.

.
.
4.2.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
[23]
.
.
.
.
.
.
.
.
.
.
.
.
.
ulated
.
.
.
.
.
.
37
.
3.3.1
.

.

.
hoice
erimen
of
.
the
.
n
.
udging
.
matrices
.
.
.
.
58
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
42
3.3.2
Burgers
Exp
.
erimen
.
tal
.

.
h
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
44
.
viscid
.
transp
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
45
3.3.3
viscid
Ph
.
ysical
.
mo
.
dels
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
46
.
Nudging
.
observ
.
[25]
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.5.1
.
ers
38
a
Lorenz
w
e
ater
quations
del
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
.
In
.
arian
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.5.3
.
v
.
study
.
a
38
simplied
Viscous
.
Burgers
.
'
.
equation
49
.

.
erimen
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.6
.
.
.
.
.
.
.
.
.
.
.
.
38
.
Shallo
.
w
.
w
.
ater
.
mo
.
del
.
.
.
.
.
.
.
.
.
.
52
.
Image
.
assimi
.
53
.
In
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
39
.
La
.
y
53
ered
Description
quasi
the
-geostrophic
[23]
o
.

.
mo
.
del
.
.
.
.
.
.
.
.
.
.
.
.
.
.
54
.
Constan
.
brigh
.
assumption
.
.
.
.
.
.
.
.
39
.
3.3.4
.
Conclusions
.
emerging
.
from
.
the
4.2.2
n
function
umerical
.
exp
.
erimen
.
ts
.
.
.
.
.
.
.
.
.
.
.
40
.
3.4
.

.

.
v
.
ergence
4.2.3
results
.
[18,
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2.4
.

.
and
.
n
.
.
.
.
.
.
41
.
3.4.1
.
Linear
.

.
.
55
.
Qualit
.
estimate
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.3
.
exp
.
ts
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
.
3.4.2
.
T
57
ransp
Sim
ort
data
equations
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
57
.
Exp
.
tal
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
42
.
Linear
.
viscous
.
transp
.
ort
..
CONTENTS
.
5
.
4.4
.
Conclusions
.
.
after
.
.
.
PhD
.
.
.
.
.
63
.
.
.
.
.
.
.
64
.
.
.
.
.
.
.
.
.
.
.
.
.
n
.
PhD
.
.
.
.
.
.
.
.
.
.
.
.
.
67
.
thesis
.
.
.
.
.
.
.
.
.
.
.
.
.
.
59
.
5
.
General
.

.
and
Publicatio
p
s
ersp
the
ecti
thesis
v
.
es
.
61
.
List
.
of
.
pu
.
blications
.
63
.
Publicatio
.
n
.
s
.
link
.
ed
References
to
theCONTENTS
6to
C

h
ev
a
broadcasting)
p

t
feature
e
made
r
of
1
sev
In
e
t
of
ro
gra-

sp
In
w
this
time
w
solving
ork,
hniques.
sev
gradien
eral
it
problems

ha
rely
v
edges
e
w
b

een
The
studied
arious
with
d
the
diusion

to
goal
y
of
t
pro-

viding
w
robust,
motiv
particularly
First,
easy
used
to
e
implemen

t,
for
fast
ob
and
image
p
o
o
the
w
ob
erful
to
algorithms.
adapt
The
d,
e
sev
ciency
segmen
of
asp
the
metho
algorithms

is
to
required
quic
b
imaging
y
or
the
the
op
If
erational
large,

or
text
th
of
fast
the
problems,
metho
tumors,
ds,
p
and
imaging
b
study
y
b
a
observ
need
top
to
is
pro


and
more
problems.
and
b
more
in
data
i
in

an
or

Ho
short
man
time.

The
the
other
iden-

subset
t
for
that

w
This
e
in
particularly
and
to
us
ok
top
in
t
to
used


t
image
is
(restoration,
the
inpain
ease
in
of
is
use
of
and
In
imple-
top
men
analysis
tation
p
of
go
the
v
metho
.
ds
er,
w
audio
e
satellite
ha
ter-
v
oth
e

dev
e
elop
pro
ed.
to
In
will


hapter
o
2,
It
w
imp
e
build
will
hemes

v
kle
real
v
of
arious
thanks
problems
more
in
o
image
erful
pro


Our
b
is
y
ated
an
y
original
eral

ations.
h
the
in
ological
the
t
eld:
generally
top
for
ological

asymptotic
design,
analysis,
shap
or
optimization
more
Also,
simply
has
the
een
top
applied
ological
electromagn
gradien
t
t.
sm
There
the
has
of

ks
tly
hidden
b

een
w
a
er,
renew
y
ed
pro
in
problems
terest
on
in
go
image
d
pro
tication

a
thanks
of
to
image,
new

applications
or
in



un

i
seemed

teresting
a
us,
t
allo
i
ed
o
to
n
the
s
ological
and
dien

metho
on
initially
one
for
hand,
k
new
to

eral
hnologie
pro
s
problems
in


tation,
un
ting).
i


teresting
a
ect
t
the
i
eed
o
the
n
d.
s
v
and
elds,
diusion
o-
of
asymptotic
information,
has
whic
it
h
ossible
no
obtain
w
o
in
results
v
ery
olv
kly
e
Ho
sending
ev
and


and
massiv
visual
e
(e.g.
o
television
ws
in
of
net
n
b
umerical
require
data
pro
(e.g.
time
images),
b
and
negligible.
on
the
the

other
is
hand
o
the
it

dela
w
the
orld,
diagnosis,
in
the
whic
w
h
data.
h
is
uge
us
progress
ortan
has
to
b
extremely
een
sc
made,
for
in
these
particular
arious
for
in
the
7
earlykw
8
assimilation,
C
sho
H
ersibilit
A
4
PTE
from
R
to
1.
at
INTR
Indeed,
O
from
D

UCTION
study
time
b
for
to
mo
e
vies
ely
and
easy
a
more
negligible
assimilation
time
of
(e.g.
ater,
smaller
made
than
ely
one
one

relaxation),
for
d.
images.
w
As
h
w
us,
e
obtain
will
ma
see
metho
hereafter,
itera-
the
b
top
,
ological
standard
gradien
terface
t
uge
metho
used
d

actually
ur-
adapts
ving
p
e

tify
to
en
image
motiv
pro
y

metho
problems,
kno
allo
obtain
wing

us
n
to
ard
obtain
it
v
kw
ery
ecause
in
ph
teresting

results
k
for
estimate
a
from
par-


the
small
forw

w

that
In
and

the
hapter
of
3,
obtained
w
b
e
assimilation
will
study
study
w
data
tly
assimilation
of
for
es-
en
e
vironmen
Ho
tal
b
and

geoph
ts
ysical
.
problems,
in
and
e
more
problem,
particularly
to
within
elo
the
images.
framew
a
ork
problem.
of
us

p
and
impro
o
the

of
observ
udging
ations.
as
F
order
or
uc
sev
results
eral
the
y
applying
ears,
metho
one

of
time)
the
noted
ma
p
jor
the

system,
has
unstable
b
the
een
of
to
problem.
appreciably
detailed
impro
3,
v
go
e
time,
our
more
kno
the
wledge
previous
of
h
these
b
turbulen
applying
t
rep
systems,
n
one
to
of
and
the
mo
ma
obtain
jor
e
goals
v
b
implemen
eing
vides
the
results
abilit
n
y
results
to
qualit
predict
are
their
uc
ev
kly
olution
using
with
ariational
a
d.
high
ts
reliabilit
the
y
these
.
elds:
Sev
images.
eral
a
dieren
tit
t
ations

images
hallenges
tially
app
impro
ear
kno
in
system
data
ev
assimilation:
images
short-range
satellites
(e.g.
v
a

few
swirls,
da
hot
ys)
ollution,
w
)
eather
ev

Sev
the

study
to
of
kind
global
w
w

arming
to
and
extract

y


hange,
ev

on
of
relativ
extreme
simple

This
phenomena
ated
sev
to
eral
the
w
ossibilit
eeks
of
in
ving
adv
of

simplest
.
ds
.
data
.
n
F
(also
or
wn
all
Newtonian
these
in
problems,
to
the
m
goals
h
are
etter
almost
without
similar.
g
They
metho

By
of
the
estimating
udging
quic
d
kly
the
and
kw
with
(in
a
problem,
v
e
ery
that
high
is
degree
ossible
of
stabilize


the
ard
state
whic
of
is
a
b
turbulen
of
t
irrev
system,
y
from
the
the
ysical

Th
bined
as
kno
in
wl-
hapter
edge
w
of

mo

dels
in
and
and
data:
a
on
reliable
one
of
hand
system
mathematic
a
a
time,
l
whic
equations

mo
y
deling
e
the
By

alternativ
atmosphere-o
and

eatedly
a
standard
n
udging
system,
d
and
the
on
ard
the

other
ard
hand
dels,
observ
e
ations
an
of
tiv
dieren
algorithm
t
is
nature
ery
(e.g.
to
in
t
situ,
pro
or
denitely
satellite
etter
observ
than
ations),
standard

udging.
onding
the
to
are
v
similar
arious
y
ph
and
ysical
often
quan
m
tities.
h
Bey
quic
ond
than
the
y
extreme
the
size
v
of
data
the
metho
problem
Chapter
to
presen
b
a
e
at
solv
in
ed
of
(sev
t
eral
o
billions
the
of
of
v
Presen
alues
,
to
h
b
quan
e
y
iden
observ
tied

from
satellite
h
is
undreds
sen
of
not
millions
to
of
v
observ
the
ations)
wledge
and
the
the
state.

w
a
er,
l
of
time
obtained
needed
y
to
denitely
solv
w
e
arious
it,

another
(h


app

ears:
of
the
w

p
of
.
dev
.
elopmen
mo
t
and
and
olving
use
time.
of
eral
a
hes
data
b
assimilation

metho
solv
d.
this
Presen
of
tly
and
,
e
it
the
is
hoice
extremely
try
dicult
iden
to
and
implemen
v
t


elds
h
the
a
of
metho
d,mak
9
as
That
elds
app
metho
eared
to
to
a
us
general
to
trix.
b
the
e
able
the
it
most
elds,
adapted
the

es
hoice
ble
for
m
rapidly
ossible
extracting
the

hniques,
v
v
en
ery
tional
ossible
data
of
(i.e.
b

these
related
data
to
h
the
in
mo
to
del

v
,
ariables),

and
it
then
guaran
b
y
eing
um.
able
these
to
e
use
extract
them

in
a
a
time,
standard
also
assimilation
pro
system.
y
The
iden
idea
h
that
view
w

e
ations
dev
ork
elop
Finally
in
and

ersp
hapter
giv
4
hapter
is
d
based
assem
on
the
the
ma-

Finally
t
a
brigh
ulti-grid
tness
h
assump-
es
tion,
p
whic
to
h
tee

qualit
of
of
lo
minim
oking
Thanks
for
all
a


w
t
are
eld
to
that

transp
elo
orts
y
an
in
image
v
to
short
another
and
one.
is
The
p
originalit
to
y
vide
of
qualit
our
estimate

the
h
tied
lies
whic
in

the
e
nonlineariza
ed
t
error
i
of
o
pseudo-observ
n
within
of
framew
the
of

assimilation.
function
some
to

b

e
p
minimized,
ectiv

are
bined
en
with

a
5.
fastODUCTION
10
INTR
CHAPTER
1.

Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin