004 
15) 
Zuler 
ntial 
(16) 
> WC 
to the 
lasses 
figure 
ve can 
show 
level) 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
  
Figure 8. Training areas Figure 9. Classified image 
The multispectral image Spot-1 of which we lay out consists of 
three channels XSi (20m X 20m), i=1, 2, 3 resulting from scene 
50-282 of February 23, 1986. The image of size 256x256 
pixels represents the area of Blida in Algeria as shown in figure 
10. 
  
  
„Rostomia-, ‚Casbah‘ ^ 
rage 
2 "Alger 
eB Le Li 
> 
    
  
   
Dousouda-Les Bai 
Douaou 
    
     
  
  
   
  
AOL TR AE 
Faraukh: 2 ; 
= Hammam Melouane 
    
   
   
  
BLIDA 
pie Tala Ai Douar 82 
„Sidi Kebír 5 Mordi_ „Tachert ; 
Bordj, sa el Habous Bel Kreit — 77 ma Hlima. ; st 
1 Ayyäch, Dayar Tádjnánt ,Chrea.- Tirhiit Ouadrar "Aut, „Oulad: Be” 
„Talaouch Ye ,Guergoür ;B Hande ^, ,Aeungal 
AUT WT ho yn S rs oF" 
Y x os ,Douar Boù Knäna ,Bahata 
: ,Quled, Brahim JMoul Aba " 8B Pouaouka Radjimi 
‘el Quail P 5 8 Annseur B Khenag 
;Tamesguid k JJarhelalet  Oulad Hammadi ,Ouled Sidi Ahmar 
c 
Figure 10. Map of North of Algeria situating the region of 
Blida. 
The classification is supervised in the sense that the algorithm 
assumes the knowledge of number of classes and their 
characteristics. The training on the satellite image allows us to 
define seven (07) classes detailed in table 11, and their 
localization is shown in figure 14. 
  
Classes | Themes 
Less dense urban zone 
Less dense natural vegetation 
Naked ground, aerodrome of 
Blida 
Non cultivated fields 
Dense urban zone (city of Blida) 
Cultivated fields 
Dense natural vegetation 
  
  
  
  
  
  
  
  
  
  
All AS tc [t32j— 
  
Table 11. The classes of the scene of BLIDA (ALGERIA) 
Figure 12. The three 
bands of the satellite 
image of Blida. 
  
  
Figure 13. Color composite of the three bands. (RGB: XS3, 
XS2 et XS1). 
  
  
  
  
Figure 14. The training areas of the satellite image. 
The resolution of equation 16 necessitates the knowledge of 
classification and restoration parameters. For our 
implementation, we have made the choice experimentally. 
Since the performance of the method depends not only on the 
choice of parameters, but it also depends on the function ©, we 
have tested the algorithm with different functions and we 
present in figure 15 the result obtained with the function of 
Hebert & Leahy. It has been proved mathematically that convex 
functions leads to convergence of criterion. The 
experimentation shows that the non convex functions may give 
better results, but the convergence is not guaranteed. 
1213