ize. 
sis for 
ion. 
  
  
  
  
  
  
  
  
  
  
1 Pixel 
  
  
  
  
  
  
  
  
  
  
16x9 9x16 
Fig.5 The effect by the size 
of the correlation window. 
  
2-2.RANDOM NOISE 
Generally, we have to take account of random 
noise included in images. Figure 6 shows the 
example of noisy stereo images corresponding to 
40dB. Figure 7 shows analyzed contours disturbed 
by noise on the deteriorated images without 
filtering. 
Figure 8 shows a result that are measured the 
error dividing between a part of top of the 
mountain and a part of foot. 
It is evident to be affected by noise in the flat 
geography like a foot area. On the other hand, the 
complicated geography is insensible to them. 
The restoring error for estimating is given by 
m m 
ZZ Mt, j)-h2i,j) | 
Error= j=li=l x 100(%) (2) 
mxmxh,s 
  
where, m is pixel size of target object, hl is the 
original height data, /2 is the obtained data and 
1, J are coordinates in the images. 
  
(b) Forward image 
Fig. 6 Noisy image. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 311