International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
following way:
= Die 7 Di
Pie TON (4)
[bir — bill
then, locates the segment orthogonal to the straight line
direction v;, and centers it at À v ;, where \ is an arbi-
trary value. The method finds the border point b;,» us-
ing the convolution with a mask, as in previous algorithm
and finally build the B-spline curve interpolating the found
points {by,...,bx}. The two first initial points b and b
are found as in previous algorithm.
5 RESULTS
A synthetic image, under a statistical model, with three dif-
ferent city-like areas and a background with was generated
simulating speckle noise. On Figure 5, the result of apply-
ing the radial straight line algorithm in a synthetic image
is shown. The result of applying the algorithm which uses
the velocity vector to find the contour of the interested ob-
ject in a synthetic image is shown in Figure 6. This image
has two regions: the object and the background. The thin
line is the initial curve and the thick line is the fitting curve
found by the algorithm. As it is illustrated, the method tol-
erates a very bad initialization step. Figure 7 (a) shows a
single look real SAR image where the algorithm was ap-
plied. On Figure 7 (b), the curves result applied to the
original image, is shown.
CAE,
NA Py EF ECS ONE
e m dmi A 2d
Me AUN ASIAN EN pe LEE
RS
Figure 5: Result of applying the algorithm of radial lines to a
synthetic image.
6 CONCLUSIONS
In this paper, a new approach to segmentation in SAR im-
ages using a classification technique based on fractal di-
mension and B-spline deformable contours, is described.
We have shown here the boundaries of several regions which
were obtained using different methods according to the
features of the object. In the first step the estimated frac-
tal dimension classification is applied to remove the noise.
The second step is to find regions of interest, in a super-
vised manner, for each region, their respective boundaries
are considered as the initial solution for the border detec-
tor. Then, a process of boundary detection is applied only
1162
Figure 6: Result of applying the velocity algorithm to boundary
extraction in a synthetic image. The thin line is the initial curve
and the thick line is the fitting curve found by the algorithm. As
it is illustrated, the method tolerates a very bad initialization step.
Figure 7: Result of applying the algorithm to a real SAR image.
for the data that are on a set of line segments. All these
processes diminish the computational cost and improve the
performance of the method. For each region, the result of
the application of this algorithm is a boundary curve given
by a mathematical formula expressed in terms of B-Spline
functions. The results using both simulated and real SAR
images are excellent with an acceptable computational ef-
fort.
REFERENCES
Blake, A. and Isard, M., 1998. Active Contours. Springer
Verlag.
Gambini, M. J., Mejail, M., Jacobo-Berlles, J. C., Muller,
H. and Frery, A. C., 2004. Automatic contour detection in
sar images. In: Proceedings EUSARO4.
Germain, O., 2001. Edge detection and localization in
SAR images: a comparative study of global filtering and
active contour approaches. PhD thesis, Université de Droit,
d'Economie et des Science d'Aix-Marseille.
Jacobo-Berlles, J., Gambini, M. J., Mejail, M. E., Muller,
H. and Frery, A. C., 2002. Bspline curve fitting in sar im-
ages. In: Proceedings EUSARO2.
Mandelbrot, B., 1983. The Fractal Geometry of Nature.
W. H. Freeman.
Int
ni
vi