Image Deconvolution by Multiscale Methods Jean-Luc Starck Service d'Astrophysique, CEA-Saclay, France.
[email protected] http://jstarck.free.fr
Collaborators: Fionn Murtagh, Eric Pantin, Mai Nguyen, Emmanuel Candes, David Donoho, Albert Bijaoui.
DECONVOLUTION METHODS IN ASTRONOMY Wiener Richardson Lucy method
Noise amplification
Maximum Entropy Method
Problem to restore stars
CLEAN Method
Problem to restore extended sources
SIGNAL PROCESSING DOMAIN
Markov Random Field
Multiscale Methods . The bi-orthogonal wavelet transfrom.
. The undecimated wavelet transfrom (à trous algorithm). . The isotropic wavelet transfrom (à trous algorithm). . The Multiscale Median Transform . New Constructions: - Ridgelet - Curvelet - Bandelet - Contourlet
The Othogononal Wavelet Transform (OWT) s l =
k
c J ,k Ë J , l k ƒ
J k
j=1
Í
j ,l
k wj ,k
Transformation H
2
C0 G c j ƒ1=
k
wjƒ 1=
k
2
h k
2l c j , k
g k
2l c j , k
C1 W1
Reconstruction: cj
,l = 2
k
– – c j ƒ 1, k h l ƒ 2k ƒ w j ƒ 1, k g l ƒ 2k
H
2
C2
G
2
W2
G H G
H
H
G
H
G
HH
GH
Smooth
Vertical
HG
GG
Horizontal
Diagonal
Undecimated Wavelet Transform
J
Isotropic Undecimated Wavelet Transform: I
x ,y =cJ x ,y ƒ
w j , x ,y j =1
J
Multiscale Median Transform:
I x ,y =cJ x ,y ƒ
w j , x ,y j =1
Multiresolution CLEAN (Wakker and Schwarz, 1998) Wavelet CLEAN (Starck and Bijaoui, 1994) Image
PSF WT
WT
CLEAN CLEAN CLEAN CLEAN CLEAN
Solution
Ideal PSF B
Reconstruction
WT
* * * * *
u ,v
FFT
º w c , j j j
B u ,v
- u ,v
2
NOISE MODELING
For a positive coefficient:
For a negative coefficient
Given a threshold t, if P > t, the coefficient could be due to the noise. On the other habd, if P < t, the coefficient cannot be due to the noise, and a significant coefficient is detected.
NGC2997 MULTIRESOLUTION SUPPORT
DECONVOLUTION SIMULATION
LUCY PIXON
Wavelet
The Curvelet Transform
J.L. Starck, E. Candès and D. Donoho, "Astronomical Image Representation by the Curvelet Transform, Astronomy and Astrophysics, 398, 785--800, 2003.
f=
i
·i Í
i
·i = f ,Í
i
Wavelet
Curvelet
Width = Length_