![](bg8.png)
(a) frame 4
(b) frame 10
Figure 4: Result with D.A. (left) and H.&S. (right).
generic form of the evolution model which will then
be specified depending of the Image Processing ap-
plication. Diffusion laws, modelling the transport of
the state vector using the physical conservative prin-
ciples, are good candidates for a number of reasons.
First, their approximation in a discrete space is robust
as stable numerical schemes exist. Second, the adjoint
of the model operator is known explicitly. Third, sev-
eral types of diffusion can be considered depending
on their impact on the spatial and temporal regulariza-
tion of the solution. An isotropic diffusion is equiva-
lent to a uniform regularization but has the drawback
to smooth the solution while an adaptive diffusion,
driven by the state vector, allows discontinuities to be
preserved. As the spatial regularization may also be
performed by the covariance matrix Q, it could be in-
teresting to compare the two paradigms. Moreover,
the diffusion can also be driven by the observation
values or by their confidence. For instance, a strong
diffusion on the state vector may operate when the
observation confidence is low which is an alternative
way to deal with missing data. Again, a comparison
between such a diffusion and the covariance matrix R
built from the observation confidence must be led.
REFERENCES
B´er´eziat, D. and Herlin, I. (2008). Solving ill-posed image
processing problems using data assimilation. Applica-
tion to optical flow. Research Report 6477, INRIA.
Hadamard, J. (1923). Lecture on Cauchy’s Problem in Lin-
ear Partial Differential Equations. Yale University
Press, New Haven.
Horn, B. and Schunk, B. (1981). Determining optical flow.
Artificial Intelligence, 17:185–203.
Huot, E., Herlin, I., and Korotaev, G. (2008). Assimila-
tion of sst satellite images for estimation of ocean
circulation velocity. In Proceedings of IEEE Inter-
national Geoscience and Remote Sensing Symposium
(IGARSS), Boston, Massachusetts, U.S.A.
Oliver, D. (1998). Calculation of the inverse of the covari-
ance. Mathematical Geology, 30(7):911–933.
Papadakis, N., Corpetti, T., and M´emin, E. (2007a). Dy-
namically consistent optical flow estimation. In Pro-
ceedings of International Conference on Computer Vi-
sion, Rio de Janeiro, Brazil.
Papadakis, N., H´eas, P., and M´emin, E. (2007b). Im-
age assimilation for motion estimation of atmospheric
layers with shallow-water model. In Proceedings of
Asian Conference on Computer Vision, pages 864–
874, Tokyo, Japan.
Papadakis, N. and M´emin, E. (2007). Variational optimal
control technique for the tracking of deformable ob-
jects. In Proceedings of International Conference on
Computer Vision, Rio de Janeiro, Brazil.
Sethian, J. (1996). Level Set Methods. Cambridge Univer-
sity Press.
Tarantola, A. (2005). Inverse Problem Theory and Methods
for Model Parameter Estimation. Society for Indus-
trial and Applied Mathematics.
Tikhonov, A. N. (1963). Regularization of incorrectly posed
problems. Sov. Math. Dokl., 4:1624–1627.
Valur H´olm, E. (2003). Lectures notes on assimilation
algorithms. Technical report, European Centre for
Medium-Range Weather Forecasts Reading, U.K.
Verwer, J. and Sportisse, B. (1998). A note on operator
splitting in a stiff linear case. Technical Report MAS-
R9830, Center voor Wiskunde en Informatica.
Weickert, J. and Schn¨orr, C. (2001). Variational optic flow
computation with a spatio-temporal smoothness con-
straint. Journal of Mathematical Imaging and Vision,
14:245–255.
VISAPP 2009 - International Conference on Computer Vision Theory and Applications
602