Figure 5: A simulated ERS-1 image of the scene shown in
Figure 1. A chi-square distributed random number speckle
file was multiplied pixel by pixel with the image from Figure
2. The number of independent looks is three.
Figure 6: This image was obtained from Figure 5 using a
(3x3) local neighborhood blurring.
olution cell. A speckle file was generated as described before
by averaging Rayleigh distributed random numbers together,
but the size of that speckle file was just k times the size of
the simulated image. The speckle file was then resampled to
the size of the simulated image, using a bilinear interpolation
transformation, and then the simulated image and the re-
sampled speckle file where multiplied together pixel by pixel.
Figure 7 was obtained by using three independent looks and
a resampling factor of k = 0.5, which is the ratio for the real
ERS-1 image.
Compared to the real ERS-1 image (Figure 1), both the blur-
ring and the resampling algorithm were found to produce re-
sults which deviate from the corresponding real image more
than the images obtained without considering the difference
between resolution cell and pixel size. This means that in the
case of our ERS-1 simulation blurring or resampling seems
not to be appropriate, because it degrades the resolution.
4 SUMMARY AND CONCLUSIONS
The principles of SAR speckle generation and its statistical
properties were briefly discussed in the introduction. Then,
we presented four different speckle simulation methods de-
scribed in the literature. Some ideas from these approaches
led to our own implementation of speckle simulation into an
existing SAR image simulator. In a practical application to
multi-look ERS-1 images, the most realistic results were ob-
tained by using a chi-square distribution on the one hand,
24
Figure 7: This image was obtained from Figure 2 by averaging
three Rayleigh distributed speckle files, and applying a bilinear
interpolation algorithm with a resampling coefficient of 0.5.
and a Rayleigh distribution with multiple file averaging on
the other hand. The application of additional blurring or re-
sampling algorithms, in order to account for the differences
between pixel resolution and radar resolution, was found to
lead to an undesired degradation in resolution.
ACKNOWLEDGMENTS
The authors would like to thank Dr. Helmut Rott from the
Institute for Meteorology and Geophysics at the University of
Innsbruck for providing the esa-ERS-1 data. Dr. Rott is Prin-
cipal Investigator of the AO/Experiment A1. We gratefully
acknowledge his cooperation.
This study was partly funded by the Austrian Academy of
Sciences.
REFERENCES
[Franceschetti, 1992] Franceschetti, G., Migliaccio, M., Ric-
cio, D., Schirinzi, G., 1992. SARAS: A Synthetic Aperture
Radar (SAR) Raw Signal Simulator. IEEE Transactions on
Geoscience and Remote Sensing, Vol. 30, No. 1, pp. 110-
123
[Goodman, 1975] Goodman, J.W., 1975. Statistical Proper-
ties of Laser Speckle Patterns. In Laser Speckle and Re-
lated Phenomena, J.C. Dainty, ed. Springer Verlag, Berlin.
[Holtzman, 1978] Holtzman, J.C., Frost, V.S., Abbot, J.L.,
Kaupp, V.H., 1978. Radar Image Simulation. IEEE Trans-
actions on Geoscience and Remote Sensing, Vol. GE-16,
No. 4, pp. 296-303
[IDL, 1994] IDL (Interactive Data Language) User Manual,
Version 3.6.1a, 1994. Research Systems Inc., Boulder, CO.
[JR, 1993] JOANNEUM RESEARCH, 1993. RSG - Remote
Sensing Software Package Graz, Software User Manual,
Release No. 3.0, Institute for Digital Image Processing.
[Leberl, 1990] Leberl, F.W., 1990. Radargrammetric Image
Processing. Artech House, Norwood, MA.
[Press, 1986] Press, W.P., Flannery, B.P., Teukolsky, S.A.,
Vetterling, W.T., 1986. Numerical Recipes. Cambridge
University Press, Cambridge, pp. 191-209
[Rainey, 1988] Raney, R.K., Wessels, G.J., 1988. Spatial
Considerations in SAR Speckle Simulation. IEEE Trans-
actions on Geoscience and Remote Sensing, Vol. 26, No.
5, pp. 666-671
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
[Wile:
nit
