Kubik Kurt 
  
image classification method introduced in Section 4.2.2, the original image can be classified to supply image blocks which 
are possible tree areas. 
The results from two steps are overlaid together and shown in Figure 8(4). Combing the results from two steps, the final 
tree areas can be illustrated in Figure 8(5). Some small areas which are extracted incorrectly, are eliminated. Figure 8(6) is 
the extracted tree areas overlaid on the image. 
is oe e S 
(1) Testing image area (2) threshold edge image(3) Tree areas from segmentation (4) Two results from Figure 6 
(5)Final delineation of tree areas (6) Extracted tree areas overlaid on image 
Figure 8 Delineation tree areas using steps in Fig.7 
  
5 TEST AND RESULTS 
Figure 9 illustrates a pair of aerial images with 630 X 714 pixels in the row and column directions respectively. The scale of 
image is 1:3437. The flying height is 519 metre. The focal length of the camera is 153mm and pixel size is 100 um . Figure 10 
illustrates the disparity map obtained from stereo image matching using robust estimation. The disparity map is further 
processed to obtain the house and tree areas illustrated in Figure 11, using the method described in Section 3. 
sm m 
     
   
   
  
    
Right — 
Figure 9 Stereo image pair Figure 10 Disparitymap Figure 11 Outlines of house Figure 12 Results of region 
and tree areas growing 
Xr. 
        
CN n | ; A s A t ra em 
Figure 13 Definition of Figure 14 Houses obtained Figure 15 The dark Figure 16 Final delineation Figure 17 Extracted roofs 
regions after the application by step 1 in Figure 3 roof area of houses overlaid on the image 
ofregion parameters 
The left image has then been processed to locate houses and to separate them from trees by the method described in Section 
4.]. Figure 12 shows the regions obtained by region growing in step 1 in Figure 3. Based on the five feature values of regions, 
most of the houses can be defined, as illustrated in Figure 13. For each region in Figure 13, the corresponding small image 
region in the original image can be identified. 
  
526 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
Bas 
can 
ext 
the 
Usi 
poi 
20 
anc 
  
Fig 
Th 
of 
are 
me 
onl 
dar 
nol 
the 
me 
Ba 
Ex! 
Bu 
Un 
Co 
Mc 
Go 
Tô 
31 
Wi 
En