International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 | Internati 
Step 3: To determine the pixels of images membership as 3. Application | 3.1. Ac 
follows: 
We tested our image segmentation algorithm on a number 
  
  
À etum of aerial images and used the results segmented for an Ino 
3 J (L.v,X)= > S (ui) D. (3) example of automatic aerial triangulation using the | photogr: 
; i=l k=l observed points grouped. principle 
1 | um. Th 
| images : 
aerial in 
trees, h 
correspc 
levels. 
reveal li 
and grot 
objects 
and grot 
image, | 
shows | 
respect 
that in 
sparse t 
some lo 
are cla: 
prelimin 
the text 
low are: 
  
  
  
  
  
  
  
  
(C) The results of preliminary segmentation (D) Final result segmented 
Figure 2. The results of aerial images segmentation Which f 
areas. In 
some lo 
where the real number me[0,x] is a weighting exponent on Nene 
each fuzzy membership. As J, is iteratively minimized, v; tie 
became more stable. Iteration terminated when 
Wa) "ia-1 € P or the maximum number of iterations is 
reached, where a. is the number of iteration and f is predefined 
tolerance. At last, high and low objects are classified 
respectively into two categories: trees and non-trees.