Keeping the highest number of differentiated information is important, 
In urban areas indeed, the heaviest difficulty arises from the large variedness 
of reflectances according to the diversity of the objects. It arises also from 
the numerous patterns characterizing a single land use, These patterns generate 
in return an equal diversity of topological structures characterizing the pixel 
reflectances. Moreover, well known in photo interpretation, the importance of 
contrast is to a bigger extent confirmed by LANDSAT data. Owing to the weakness 
of pixel ground resolution, some rather small sized objects will not appear but 
thanks to their highly differenciated reflectance from surrounding objects in one 
or several channels. 
Other objects, even bigger sized than the pixel, will not be weel dif- 
ferenciated. This is due to their low level of contrast. The object recognition 
will thus depend on its surroundings. It follows that the land use scarcely ap- 
parent in the image will be unlikely identified in such an environment. 
These statements lead to two kinds of consequences : 
1) Urban areas interpretation, as stated before, calls for a maximally 
diversified information. 
2) It relies more on the recognition of reflectance values (or of 
their components) relevant to urban structures than on the analysis at the pixel 
level of rough spectral signatures of detached objects. 
In this order a method has been worked out using the nearly total 
(98 Z) multispectral information and enabling the factorization of numerous clus- 
ters. 
A next step will consist in distinguishing by means of a human assis- 
ted analysis the signatures proper to specific urban land uses. Behind this ap- 
proach which is not always feasible, a possible method exists of discriminating 
between different land uses. This method implements spatial grouping of spetral 
signatures at certain rates in a given area on the image, in order to point out 
a specific land use or a combination of several land uses covering the related 
area. 
      
  
  
  
  
  
  
  
  
  
  
  
   
  
  
  
  
  
  
  
  
  
  
  
  
  
   
  
  
  
  
  
   
  
   
   
hard 
prog 
the 
and 
enat 
bot! 
intc 
Besi 
urb: 
cat; 
opei 
pea 
thi: 
tan 
and 
C OIX