SAR SPECKLE SIMULATION 
Regine Bolter, Margrit Gelautz, Franz Leberl 
Institute for Computer Graphics 
Technical University Graz 
Münzgrabenstrafe 11 
A-8010 Graz, Austria 
e-mail: bolter@icg.tu-graz.ac.at 
Commision Il, Working Group 4 
KEY WORDS: Remote Sensing, Simulation, SAR, Speckle 
ABSTRACT 
After a short introduction to the principles of SAR speckle generation and its statistical properties, we give a review of different 
speckle simulation methods described in literature. Then, the implementation of some selected algorithms is described, and 
their performance is tested on simulated ERS-1 images. Special attention is paid to the modeling of multiple looks, and the 
differences between image pixel size and original radar ground resolution. A chi-square distribution and a Rayleigh distribution 
with multiple file averaging were found to produce the most realistic results. 
KURZFASSUNG 
Nach einer kurzen Einführung über die Prinzipien der SAR-Speckle Entstehung und einer Beschreibung der statistischen 
Eigenschaften geben wir einen Überblick über verschiedene Speckle-Simulationsmethoden, die in der Literatur beschrieben 
sind. Dann wird die Implementierung einiger ausgewählter Algorithmen beschrieben, und ihre Performance wird anhand von 
simulierten ERS-1 Bildern geprüft. Dabei wird besonderes Augenmerk auf die Modellierung von Multiple Looks, sowie der 
Unterschiede zwischen Bildpixelgröße und ursprünglicher Bodenauflösung des Radars, gelegt. Die Verwendung einer chi-square 
Wahrscheinlichkeitsverteilung, sowie einer Rayleigh-Verteilung mit Mittelung über Mehrfachfiles, lieferten die Ergebnisse mit 
der größten Realitätstreue. 
1 INTRODUCTION square area on the ground. Creating a multiple look image 
reduces the speckle effect, but at a cost of reduced azimuth 
SAR (Synthetic Aperture Radar) simulation is an important com, 
tool for the development and testing of SAR image processing 
algorithms, since simulation provides inexpensive and flexible ~~ The interpretation of speckle as a random effect leads to sta- 
test material, which often cannot be obtained from other tistical descriptions as given by, e.g., [Goodman, 1975]. The 
sources. Furthermore, in those cases where no ground truth magnitude of the resultant speckle field follows a Rayleigh 
data is available, e.g. planetary mapping, simulation is often probability distribution, the phases are uniformly distributed, 
the only means of verification. In order to make the simulated and the speckle intensities can be described by a negative ex- 
imagery look as realistic as possible, the proper treatment of  ponential distribution. The analysis also shows that N non- 
SAR speckle noise is an important issue. This is to avoid coherent "look-averaging" results in a density function for 
situations where image analysis algorithms perform well on intensity that is chi-squared with 2N degrees of freedom. As 
simulated data, however fail in their actual application to N increases, this distribution approaches a normal distribu- 
real SAR data. tion. 
Speckle is a physical effect, which occurs when coherent light ~~ When dealing with SAR simulation, we have to distinguish 
is reflected from an optically rough surface. In remote sens- between raw signal simulators and SAR image simulators. In 
ing SAR sensors, speckle results from the need to create the raw signal simulators, first the signal received by the sensor 
radar image with coherent radiation. A single resolution cell is generated, and then this signal is passed through an ap- 
typically covers a 25m x 25m area on the ground. The trans- propriate Doppler processor to produce the final SAR image. 
mitted wave length onto that resolution cell is only a few Speckle is considered at signal level, by simulating the phys- 
centimeters. Due to the texture and roughness within a res- ical process of adding up the different scatterers falling into 
olution cell numerous targets exist that produce the radar one resolution cell. A prerequisite for this technique is a high 
echo. The characteristic speckle effect of radar images results resolution Digital Elevation Model (DEM) which contains in- 
from the destructive and constructive interferences among formation about the surface micro structure. In SAR image 
the echoes of individual surface scatterers within a resolution simulators, information about the DEM and the imaging ge- 
cell. Therefore the resultant pixel can differ extremely from ometry is used to compute the gray values of the SAR image 
its average gray value. These gray value variations between directly, without intermediate signal representation. Speckle 
adjacent pixels lead to the typical grainy appearance of SAR is added to the final image by appropriate modeling of its sta- 
images. tistical properties. The absence of speckle in radar shadows 
: : needs also be taken into account. 
The nature of speckle can be interpreted as a random effect. 
This in turn can be reduced by averaging adjacent pixels. In the next section, several SAR speckle simulation techniques 
If the resolution in range direction is lower than in azimuth presented in the literature are reviewed. Afterwards, we de- 
direction, a multiple look image can be created by averaging scribe and discuss the implementation of some selected algo- 
several image range lines so that the resulting pixels cover a rithms, which were embedded into an already existing SAR 
20 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
imag 
ERS- 
ison 
Whe 
specl 
only 
An e 
imag 
when 
nale 
Is mc 
simu 
dictk 
elem 
bem 
least 
After 
the c 
the \ 
obta 
form 
used 
to tt 
from 
wher 
two 
by G 
flecti 
incid 
tion 
from 
Spec 
looks 
wher 
resol 
squa 
is th 
inde 
wher 
unit 
tion 
the ç 
wher 
is th 
signe