Evaluation of Several Speckle Filtering Techniques for ERS-1&2 Imagery 
Yonghong Huang and J.L.van Genderen 
Department of Earth Resources Surveys 
International Institute For Aerospace Survey and Earth Sciences 
7500 AA Enschede, The Netherlands 
Tel: +31,53,4874254 Fax:+31,53,4874336. 
KEY WORDS: Speckle Noise, SAR Speckle Reduction, Filtering Techniques. 
ABSTRACT: 
The speckle noise appearing in SAR images forms a main obstacle to analyse, interpret and classify SAR images for 
various remote sensing applications. To date, many filters have been developed for speckle reduction in SAR 
imagery. In this paper, the authers evaluate eight ready-made speckle filters, which are Moving Average Filter, 
Median Filter, Lee Filter, Enhanced Lee Filter, Frost Filter, Enhanced Frost Filter, Kuan Filter and Gamma MAP 
Filter, in order to test these filters with ERS-1&2 data and provide a guideline for users are interested in SAR 
applications. The evaluation has been done from three aspects, which are statistical model testing, quantitative 
evaluation and qualitative evaluation. The test results presented in this paper confirm that filtering techniques are 
helpful for SAR image applications. 
1. Introduction 
The earliest concept of speckle is introduced from the 
laser field[1], which indicated that the interference of 
the coherent and dephased reflected scatters will cause 
a peculiar granular pattern in the image due to the 
majority of surfaces being extremely rough on the scale 
of the wavelength. This concept is suitable to SAR 
speckle as the SAR system has a similar principal with 
the laser imaging system. Therefore, the SAR speckle is 
related with laser speckle both physically and 
mathematically so that the large amount of material on 
laser speckle is applicable to understand the 
phenomenon of SAR speckle. 
For several decades, many mathematical models of 
SAR speckle noise have been deeply investigated 
[2][3], which took the speckle noise as a kind of 
multiplicative noise. By using the statistical properties 
of the SAR image, the probability density function(pdf) 
of a normalized multiplicative fading process can be 
expressed as an exponential pdf, which provide the 
fundamental theory for SAR speckle reduction 
techniques. Many speckle filtering techniques based on 
the speckle model of multiplicative noise are well 
developed. 
In solution to the incoherence of SLAR systems, which 
are relatively immured to speckle due to performing an 
incoherent summation of N independent estimation of 
the radar backscatters with the same resolvable 
element, the multi-look algorithm[4] has been proposed 
to reduce the speckle noise. It is realized in the 
frequency domain by segmentation of the azimuth 
164 
spectrum to form several independent single-look 
images and summation of these images to obtain a 
multi-look images. The drawback is that the speckle 
reduction is accompanied by loss of image resolution. 
The second category technique to reduce speckle noise 
is referred to the time domain algorithm, as it performs 
speckle filtering after a full resolution is processed. It 
has advantages over the conventional frequency domain 
multi-look technique due to the fact that it is much 
easier to compromise the balance between the image 
resolution and speckle reduction, and more flexible to 
filter the full resolution images with the different filters. 
Many speckle filters are well developed, such as Box 
Filter[4], Median Filter[5], Lee Filter[6], Enhanced Lee 
Filter[10], Frost Filter[8], Wiener Filter[11], Kuan 
Filter[11], GMAP[9], Geometric Filter[13] and so on. 
The main goal of these speckle filters is to reduce 
speckle noise to a minimum, in order to achieve image 
quality as good as photography does. However, these 
filters still cannot meet the requirement of the purposes 
of easy interpretation and clear classification. The 
reason is that the properties of SAR speckle are more 
complicated to be modeled as an exponential 
distribution. In some cases, the statistical model 
violates this distribution. The different terrain and 
different SAR systems, for example, terrain size, terrain 
geometry, moisture, dielectric constant, wavelength, 
polarization, view angle and so on, will generate the 
different speckle noises. How to overcome it with one 
filter has become a difficult and long-term task that has 
been treated by many scientists. In spite of these 
deficiencies discussed above, the majority of image 
processing software packages have been formed 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
CN ©) DD MB 0. 7] 2 ©) OY M rp AN 40 o o ©) oM M y E o CS). 05. OY A o; 6 
a 
iummd 
IM cg " mes pugs AW 9 —— BO 
— — ee 
"PN peed pede