3. Istanbul 2004 
in figure 5 to the 
tation (figure 4) 
| shadow area are 
lered as potential 
and orientation 
ibute permit also 
ation 
od 
recovery method 
use as input : the 
nents and their 
and the original 
computing. 
Definiens, 2002) 
ntropy, contraste, 
ent with a texture 
ed. The result is 
data. Among the 
nity give the best 
1ap. 
resented at figure 
exture as shadow 
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
Some difficulties for calculating the segment texture are related 
to the size and form of some segments. Few segments are very 
small and texture on that segments is not significant or is not 
well computed. Other difficulties are related to the texture at 
segment borders and the heterogenety inside segment due to the 
very high spatial resolution. 
The shadow effects compensation on image is done on each 
shadow segment. The gamma compensation parameter is 
calculated for each shadow segment and all pixels in shadow 
are corrected using the gamma parameter and its value in 
shadow. Some results of shadow compensation are presented on 
figure 7 . The de-shadowed images are visually untached. The 
only drawback is on the shadow transition area between the 
shadow and other surounding surfaces. 
  
Figure 6 :Neighbouring segments with the same texture as 
shadow segment using the contraste texture feature 
6. CONCLUSION 
The contribution of the contextual and geometric attibues in the 
shadow detection method is very important. The results are a 
precise buildings shadow detection on the panchromatic Ikonos 
images. 
The information under shadow retrieval is quite possible using 
the contextual and texture attributes of shadow and its 
neighbouring segments. The results are very promizing and 
have a great potential of application to correct shadow negative 
effects on images. Results can also be used to complete land use 
map derived from the very high spatial resolution images 
To increase the information retrieval precision, more 
investigations on the texture attributes computing are 
recommanded. 
ACKNOWLEDMENTS 
This work was supported by the NSERC (the Natural Sciences 
and Engineering Research Concil of Canada) and the PCBF 
(Progamme Canadien de Bourses de la Francophonie) of CIDA 
(Canadian International Development Agency) 
  
  
  
Figure 7 : Shadow and de-shadowed images 
(a) . Blocs 1-7 image 
with shadow 
(b). Blocs 1-7 
deshadowed image 
(c) Sciences Faculty 
Bloc image 
(d). Sciences Faculty 
Bloc de-shadowed 
image