Istanbul 2004 
niques with 
help airfield 
his paper, we 
obstructions 
the Airfield 
obstructions 
manager can 
s from the 
an automatic 
‘his will meet 
identification 
ed number of 
rinciples of 
sity Press. 
1 of Control 
ogrammetric 
;. 
- information 
SPRS Annual 
ligital terrain 
. Proc. of the 
p. 
g: A Remote 
echniques of 
p. 
Information 
ation Safety, 
informatics, 
01. Airfield 
ip. 
s: Planning, 
AcGraw-Hill 
technologies 
December 
> 
Dr. Richard 
niversity of 
latasets. The 
nsortium. 
USE OF STATISTICAL DISTRIBUTION FOR SEGMENTATION 
OF SAR IMAGENS OF OCEANIC AREAS 
R.F.Rocha 
Remote Sensing Division, Hydrographic Navy Center, Av. Baráo de Jaceguay s/n, 24048-900, 
Niteroi, RJ, Brasil - renatofrocha@hotmail.com 
KEY WORDS: SAR, Statistics, Ship Detection, Processing, Image, Algorithms, Oceans, Remote Sensing. 
ABSTRACT: 
In this work the use of statistical techniques will be approached for segmentation of SAR images, with the purpose of ship detection, 
being used RADARSAT images of the Brazilian coast. As described in Rocha et al (2001) and Rocha and Stech (2003), a specific 
software for ship detection was developed where, based on Eldhuset (1996), Vachon et al (1997), Oliver and Quegan (1998), Zaart 
et al (1999), Ferreira et al (2000) and Macedo et al (2001), some routines were implemented for the use of several statistical 
distributions for the segmentation of the images, as, for example, the Weibull, Gama, and K distributions. Initially, the shape and 
scale factors were dear in function of the statistical characteristics of each image, as average, variance and standard deviation. 
However due to the variability of these characteristics in agreement with each image, they were established values patterns for these 
factors, that allowed a desirable adaptation of the curve of the distribution to the curve of the histogram of the image. These routines 
were, then, tested for this group of images and its results were analyzed. The results were analyzed individually, through a 
comparison among them and, also, using a RGB composition among them. 
1. INTRODUCTION 
Brazil possesss a vast territorial sea and the necessity of a 
control on targets if it makes indispensable. Diverse reasons 
motivate us to develop technologies directed toward the ships 
detection, as cited for Fingas and Brown (2000), such as the 
combat to the illegal fishing, the combat to the drug traffic and 
the control of our territorial sea. 
Initially created for military use, diverse targets detection and 
monitoring systems had been developed, being called 
Automatic Targets Recognition systems (ATR). In the last 
years satellites images have been used for the monitoring and 
the detection of oceanic targets and boats. McDonnel & Lewis 
(1978) demonstrated that the ship detection of great load is 
theoretically possible in the visible bands with LANDSAT 
images. However, due to problems of clouds covering and low 
temporary repetitivity, this methodology was not made possible 
operationally. 
With the development of the technology of the synthetic 
aperture radar (SAR) and the appearance of satellites using this 
technology, an growing number of works has been developed in 
the most several areas of performance in the last years. Such 
fact made with that innumerable techniques for the SAR images 
processing were developed, in order to allow a convenient 
extration of information, according to the area of interest. Due 
to the inherent characteristics its formation, SAR images have a 
factor that make difficult its interpretation, that is speckle, 
whose reduction, either total or partial, besides difficult, it can 
cause the loss of some information. Staples ct al. (1997) 
demonstrated the viability of the use of RADARSAT SAR 
Images for ships detection. The results of this study indicated 
that the detection of ships with SAR images depends on a series 
of factors, such as: direction and intensity of the wind, size of 
the ship, route of the ship in relation to the direction of sought 
of the radar and the direction of aimed at of the radar in relation 
to the direction of the wind. 
Le) 
CA 
2. METHODOLOGY 
The area of study focused in this work is situated in the area of 
Santos (SP), one of most important and busy harbor of Brazil. 
It was used in this work a RADARSAT SAR image, Standard 
mode, acquired on December 17, 2003. 
Amongst the several referring existing works to this theme, we 
base our work on the methodologies described by Eldhuset 
(1996), Vachon et al (1997), Oliver and Quegan (1998), Zaart et 
al (1999), Ferreira et al (2000), Fernandes (1998) and Macedo 
et al (2001). All these works are base on the use of statistical 
distributions for segmentation of SAR images. Amongst these 
diverse statistical distributions, K distribution has been used as 
a flexible tool for modelling of data deriving of a SAR image, 
as described by Yanasse et al (1994). In the case of SAR images 
of oceanic areas especially, Vachon et al (1997) corroborates 
the statement of Yanasse et al (1994) and uses K distribution for 
ship detection. However, as described in Oliver and Quegan 
(1998), other statistical distributions can also be used in the 
SAR image processing. According to this author, the Rayleigh, 
Weibull and Gama distributions can also be used for the SAR 
image processing, depending on the area of study. Some other 
authors also demonstrated the adaptation of these distributions 
for the SAR image processing. The distribution Gamma is also 
used for the segmentation of SAR images for Zaart et al (1999). 
Ferreira et al (2000), Fernandes (1998) and Macedo et al (2001) 
demonstrated the use of the Weibull distribution in the 
segmentation of SAR images. Already in the method 
demonstrated by Eldhuset (1996) a new image is generated after 
the use of some statistical parameters, as mean and variance. 
The detection of possible targets in SAR images is 
accomplished by means of the great difference of values of grey 
level presented by ships in contrast with the water. In general 
way, in a SAR image the ships are present as points with high 
values of grey level, while the water is presented as points with 
low values of grey level, except for the pixels of water strongly 
contaminated by speckle. However these basic characteristics