CNES research project
Image
restoration
The
problem of recovering an image that has blurred and corrupted with an
additive noise is a classical problem in image processing.
There
is no really agreement on an optimal solution.
Among
the methods that have been proposed to solve this problem, two of the
most successful are that of Total Variation (TV) image restoration
proposed by L. Rudin, S Osher and E. Fatimi in their seminal work and
the thresholding of wavelet coefficient decomposition (cf. D. Donoho,
Y. Meyer
).
These
two kinds of methods are quasi equivalent up to a different handling.
TV
restoration requires the choice from all the TV images that is
smallest: TV measures the discontinuity weighted by its jump. This
hypothesis is not reasonable to recovering texture areas. VT
restorated images are corrupted with "pixellised"
phenomenon on homogeneous areas. It is not observed with thresholding
method but the impulse response of wavelet is visible on the images.
In order to manage this thresholding and to take into account edges,
frame geometric approach has been proposed. S. Mallat (bandelet
decomposition) proposes an adaptive method and E. Candes a systematic
one (curvelet decomposition).
The
project will consist in tutorial on satellite images. Results will be
analyzed and we will look for theoretical explanation (image
regularity, TV or Besov image representation space, optimality
theorem etc.).
The
aim is to gather researchers and engineers who are interested in all
restoration image aspects.
Participants:
Andrès Almansa,
Sylvain Durand,
Yves Meyer,
Jean-Michel Morel,
Mila NiKolova,
Bernard Rougé.
|