378 
THE INFORMATION CONTENT OF RADAR IMAGES, 
MODELLED ACCORDING TO CORRELATION PROPERTIES OF THE SIGNAL. 
R. Okkes 
European Space Research & Technology Centre 
Noordwijk, the Netherlands 
Comission II 
1. | INTRODUCTION 
By treating radar imagery data extraction as a communication system in 
which the target information is the source and the communication channel 
is modelled according to the noise properties of the radar sensing 
process, the information content of the image can be expressed in terms of 
the average information rate which can be transmitted through that noisy 
channel. In this paper the information rate is derived using fundamental 
theorems of information theory for a commonly used model of the target 
statistical properties, which allows to gain insight into radar image 
formation effects like "look" summation and radiometric resolution. It is 
shown that the information content per spatial resolution element of the 
image is small for a low number (e.g. four) of looks. As a measure of 
image quality an alternative definition of radiometric resolution, based 
on rate distortion theory, is introduced which takes account of image 
correlation properties. Similar derivations has been performed by Frost 
et al for uncorrelated imagery (ref. 3). 
2. SAR IMAGE STATISTICAL PROPERTIES 
  
2.1 Image Pixel Statistics 
The statistical properties of speckled SAR imagery of homogenous areas can 
be conveniently expressed by writing the observed pixel intensivity (I.) 
in the form of: } 
rs. 1! 1 
i 421 (1) 
where: 
5. = true (mean) reflectivity of the area - 
L' » number of independent samples of intensity averages (looks) to form [, 
n = random variable describing the speckle. 
The propability density function (pdf) of n, assuming power detection is 
given by: 
P(n) s got amt for .n > 0 (2) 
: (L-1)! 2L d : 
which is a chi-square distribution. Note that expression (1) separated 
the observed intensity into: 
a) a mean value component and 
b) a signal independent multiplicative noise component. 
The mean and variance of the component n are respectively: