Area TM SPOT IRSx 
Left Right India 
France 22/5/88 31/10/89 | 17/10/89 
(south) 1A,Pan 1A,Pan 
Vertical | L17°.2 R26°.5 
30 m 10 m 10 m 
Hannover [27/4/87 (17/6/86 28/6/86 
(Germany) 1A,XS 1A,XS 
Vertical | L8°.6 R8°.0 
30 m 20 m 20 m 
New-Delhi 12/5/86 13/11/88 
(India) 1B, XS 
R2°.3 Vertical 
20 m 36 m 
  
  
  
  
  
  
  
* - linear array sensor 
Table 1 Landsat-5  TM,IRS and SPOT scene 
characteristics (acquistion date, product, vieving 
angle, resolution) 
IMAGE PREPROCESSING 
Image preprocessing deals with the preparation of 
diapositives to be used on the analytical plotter. 
For multispectral images first a choice had to be 
made by of the proper combination of bands to 
produce a black and white diapositive with the 
greatest information content and the least 
correlated bands. After various trials, bands 
2,4,5 and 2,3,4 were selected respectively for TM 
and IRS images. The usual image enhancement 
techniques were applied in order to obtain a 
"sharp" image and to facilitate the interpretation 
and extraction of linear features. Several 
operators were tried out on the Context Vision and 
finally the Wallis operator was accepted for 
giving the best result for TM and SPOT images, 
based on visual inspection. 
The next step of the image processing dealt with 
the geometric image transformation. The oblique 
SPOT image (usually level 1A) was kept as the 
reference image. The selected image (here TM or 
IRS) was geometrically transformed into the SPOT 
image, with the help of some common reference 
points. 
Prior to the resampling, the TM and IRS images 
were scaled to the higher resolution SPOT image; 
it was expected that this would lead to reasonably 
good image quality in the resampled image (Tauch 
and Kaehler 1988). 
the selection of 
Special care is required for 
reference points, which is not an easy task, due 
to the different spatial resolutions of both 
images. Relief displacement in the oblique SPOT 
of 500 m can lead 
within scan lines, 
image for height differences 
to shifts of several pixels 
depending on the viewing angle. 
It was therefore important to select the reference 
points at the same height level. An affine trans- 
formation was applied to the six common reference 
points distributed along the edges of the 
overlapping images. 
304 
Results of these transformations are summarized in 
table 2. 
  
  
Data set | Reference |Transf. RMSE 
image image (pixel) 
1 SPOT Pan TM 1.0 
R 26°.5 
2 SPOT XS TM 0.4 
L 8°.6 
3 SPOT Pan IRS 1.0 
R 22.93 
  
  
  
  
  
  
Table 2: Image transformations 
During geometric transformations and subsequent 
resampling, some distortions may be introduced to 
the transformed image. In order to check the 
magnitude of these distortions a reseau of 25 
cross marks at a regular spacing of 250 pixels was 
created in the original TM image. These points 
were measured on the analytical plotter DSRI 
before and after the geometric transformation and 
resampling. Applying an affine transformation 
using 6 control points with nominal coordinate 
values, we obtain the following results: 
Original TM: RMSExy = 0,017 mm or 6.8 m 
Transformed TM:RMSExy = 0,018 mm or 7.2 m 
(image scale: 1:400 000) 
This shows that there is no significant distortion 
as a result of the image transformation. But on 
the other hand, geometric errors introduced during 
the whole conversion process from analogue to 
digital’ are not negligible. Fiducial marks are 
usually not available on satellite images, 
although they would be useful even for SPOT 
images. 
When we use transformed images like TM or IRS, 
they overlap only partly with a SPOT image. It is 
therefore necessary to create artificial fiducial 
marks defining a square of the same size as a SPOT 
image. 
EVALUATION OF PLANIMETRY AND HEIGHT ACCURACY 
Multisensor stereo data sets are recorded by two 
different scanner systems, from different orbit 
paths and different heights. This leads to a low 
B/Z ratio which greatly influences the height 
accuracy. 
Various mathematical models 
for the 
analytical 
popular 
have been developed 
restitution of satellite images on 
plotter (Konecny 1987); the most 
are the orbital parameter model (Guichard 
1984, Gugan, 1988) and the extended collinearity 
model. Our experiments were carried out on a DSRI1 
analytical plotter, using a SPOT software suite 
developed by the Joanneum Research Center, Graz, 
Austria. The software uses the extended 
collinearity model, where the positional and 
attitude parameters are time dependent and can be 
modelled by polynomes, linear or 
quadratic changes. 
allowing for