AUTOMATIC TIE-POINTING IN OVERLAPPING SAR IMAGES 
Arnold BAUER, Hannes RAGGAM, Wolfgang HUMMELBRUNNER 
Institute for Digital Image Processing, JOANNEUM RESEARCH 
Wastiangasse 6, A-8010 Graz, AUSTRIA 
Commission Il, Working Group 11/4 
KEY WORDS: SAR, Matching, Software Development, Feature Extraction, Image Understanding 
ABSTRACT: 
Within the German ground segment for ERS-1 SAR data the geocoding system GEOS has been developed for the 
operational generation of geocoded ERS-1 products. Among other contractors, the Institute for Digital Image Processing 
has contributed essential modules to this system. With regard to individual ERS SAR scenes, these mainly cover the 
set-up of SAR mapping parameters, the optimisation of these parameters using least squares adjustment techniques 
and tools for a meaningful quality control of geocoded products. Recent developments have been related to the 
automatic detection of tie-point candidates in overlapping SAR scenes being acquired either along one satellite track or 
from adjacent satellite orbits. Therefore, in a first step candidate points, i.e. points showing very specific features, shall 
be automatically detected in the respective overlapping areas of a selected SAR reference scene. In a second task the 
corresponding points have to be detected in the areas of the other image(s) overlapping the reference image. This is 
based on automatic image matching techniques. Potential application areas for such points are given for instance 
through SAR image coregistration and SAR block processing tasks, respectively. 
certain operations included in the processing. The 
algorithms and methods, which have been implemented 
for the detection of tie-point candidates in SAR images, 
are presented in detail. 
1. INTRODUCTION 
Since the launch of the European ERS-1 sensor, SAR 
images are available on a continuous basis. Meanwhile 
also ERS-2, the Japanese ERS-1 and the Canadian 
RADARSAT are in orbit and SAR data become 
increasingly used for various applications. With regard to 
the ERS ground segments the Institute for Digital Image 
Processing is involved in the German PAF and 
contributes to the development and maintenance of the 
geocoding system GEOS, which has been developed for 
the operational generation of geocoded ERS-1 products. 
In this concern, recent developments at the Institute have 
been related to the automatic detection of tie-point 
candidates (TPCs) in overlapping SAR scenes. Such 
points may be particularly useful for the geometric 
modelling and processing or for coregistration of multiple 
SAR images including stereo models. 
Automatic image matching techniques are used in order 
to find the corresponding points of the detected tie-point 
candidates in the overlapping areas of the other image(s). 
Therefore, a technique based on image feature vectors is 
used, which makes use of particular derivatives of a SAR 
image, so-called features, being created by filtering the 
image with methods as mentioned above. The perfor- 
mance of the matching algorithm is analysed for those 
points provided by the automatic tie-point detection 
procedure. 
2. TIE-POINT CANDIDATE DETECTION 
The tie-point candidate detection procedure aims at the 
automatic definition of characteristic points and regions, 
which presumably can be detected in the overlapping 
1. candidate points in a reference scene, i.e. points area of another SAR scene with high probability. In the 
showing very specific features, shall be automatically ^ ideal case, man made features like streets or railroads or 
detected in the respective overlapping areas of the crossings of such linear features should be typically 
SAR image; detected. However, such features may either not always 
be available or may not be clearly visible in SAR images 
due to the SAR speckle noise. 
In general, the tie-point detection has been designed as a 
two-step procedure as follows: 
2. the corresponding points in the areas of the other 
image(s) overlapping the reference image shall be 
found automatically using proper image matching 
techniques. Moreover one has to cope with other SAR specific 
peculiarities like layover, foreshortening and so-called 
Strong scatterers. Therefore, respective preprocessing 
steps like prefiltering or preclassification have been 
implemented in order to increase the detectability of 
characteristic points in a SAR scene. For the final 
Emphasis in this paper is put onto the developments 
related to the automatic detection of candidate points in a 
SAR image. Experience has shown, that this is a rather 
complex task as SAR images are particularly sensitive to 
315 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996