C 
  
the 
ints 
SM 
ing 
oth 
the 
ent 
olar 
olar 
International Archives of the Photogrammetry, Remote Sensing 
images, in fact, increases the speed of the correlation process. 
Once the images have been registered to the same ground area, 
positional differences are assumed to be due to parallax, which 
results. due to relief. Measured parallax differences are 
converted to absolute elevations using trigonometric functions 
and the orbital data (orbital position, altitude, attitude and the 
scene center). 
The automated image matching procedure used to derive the 
elevation from the parallax and produce the DSM is carried out 
through a comparison of the grey values of the two images. This 
procedure is based on a mean normalized cross-correlation 
matching method with a multi-scale strategy to match the image 
using the statistics collected in the defined windows. The multi- 
scale strategy is based on a hierarchical approach using a 
pyramid of reduced resolution images. The first attempt at 
correlation is performed on very coarse version of the images. 
This enables the software to match the features more accurately 
and, subsequently, improves the correlation success. The next 
correlation attempts are performed using finer features and on 
higher resolution versions of the image. The full resolution is 
then used in the last correlation. 
The result of matching procedure is represented by correlation 
coefficients, varying between 0 and | for each pixel, with 0 
representing a total failure of the matching and 1 a perfect 
match. A refinement of the matching procedure could be 
performed in order to improve the accuracy to sub-pixel level. 
Elevation points are extracted at every pixel for the complete 
stereo pair. As previously said the advantage of using this 
procedure is that the search for the matching pixels is limited to 
the epipolar line, allowing a significant improving of the 
algorithm efficiency and accuracy. 
The percentage of the correlation successfully performed was 
around the 99%. The figure 4 represents the correlation index 
computed over the whole image using a grey level scale. 
  
and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
Figure 4. Correlation index plotted as grey levels; white pixels 
indicate the full correlation whereas black pixels indicate a low 
correlation or correlation failure 
The grid spacing of the extracted DSM has been selected to 2 
pixels, corresponding for the EROS images processed, to 
approximately 4.6 m. The resulting file is composed of more 
than 8 million of points with an elevation ranging between 15 
and 490 m a.s.l. In figure 5 a perspective view of the geocoded 
DSM. 
  
  
Figure 5. Geocoded Digital Surface Model. Elevation. Coordinates are expressed in the National - Gauss-Boaga grid. 
Holes, depending on the failure of the correlation phase, are visible in figure as white area. 
1051