Figure 2. a) Landuse map in GIS b) Color composite of the 
bands 5, 4, 3 of TM (RGB 5, 4, 3) 
(size is 215 , 340 pixels) 
c) MLH classification result — d) A zoomed part of the 
classified image shown in Figure. 2(c) 
  
Figure 3 Two samples of the generated likelihood maps for the 
selected classes. 
a) Likelihood map for class Beans 
for class Potatoes 
b) Likelihood map 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
   
  
     
   
   
      
    
  
    
      
    
Polygon 
Boundaries 
  
Likelihood map 
prob(Cid) 
   
  
Overlay of the 
polygon map 
& 
Likelihood map 
Extracting the 
average 
  
  
probabilities for 
each polygon 
Thereshold 
ing 
Binary t 
hypothesis 
  
       
  
   
  
  
Hypothesis 
evaluator 
  
  
  
   
   
    
    
  
Shape 
hypothesis 
generation 
  
Figure 4. Schematic diagram of the proposed method for MBIA 
of the agricultural areas using the existing boundaries. 
On the other hand we need the evidence maps to calculate the 
cost. In the first stage, we had generated the likelihood maps for 
each class individually. Then if we assume that we have just 7 
classes in the image as we used for this data we can calculate 
the evidence map for others using a similar formula as it was for 
hypothesis maps namely: 
B2=1-El (3) 
In which El is the evidence map for that particular class which 
we are working on it. Performing these calculations now we 
have all of the components that we need to estimate the cost of 
the specific threshold t; therefore using equation (1) we 
calculate the cost. After this we examine that this cost is 
minimum or not. If it is not, then we change the threshold value 
t to t' and repeat the whole procedure for the new hypothesis 
map. The cost graph of the various thresholds for class Onions 
have been shown in the Figure 5. Table 1 shows the 
corresponding minimum cost for each class. We store the final 
result for each class to merge them in the final map by assigning 
the relevant IDs to each resulted hypothesis map. 
Integration of all of the final hypothesis maps can be seen in the 
Figure 6(b). In this map each polygon has a label defined by the 
procedure. As it has been shown some unlabeled polygons are 
in the final map. This implies that these polygons could not be 
categorized into any class. In other words, for all of the classes, 
the probabilities have not supported any of them. These 
polygons are shown in the Figure 6(c). 
1284 
Internationc 
Cost 
Figure © 
Onions ¢ 
  
Table 1. T 
As it is cle 
crop type ( 
that the pr 
that many 
be sure th: 
assumption 
polygon th 
polygons « 
boundaries 
into the ne 
polygons I 
caused the 
of these ern 
that the po 
indeed som 
fall into thc 
been shown 
5 
À common 
images is O 
parcel-base« 
this method 
each pixel i 
existing boi 
each polygc 
majority in