The matching point is a point which has largest 
correlation coefficient 
N1-1 N2-1 a CT B (1) 
C(a,b)= S s (1 b)y(m,n) - I)(T(m,n) - T) 
  
  
m=0 n=0 10 To 
where I 1 5 
= —— I(a,b)(m,n) , 
NixN2, "7 dao 
T 1 NS 
= T(m,n) , 
NIx N2 -0n=0 
N1-1N2-1 
[Cab = $ > {1(a,b)(m,n)-1}? : 
m=0n=0 
N1-1N2-1 ; 
Tes. S «X offimn)-Tj: 
m=0n=0 
And the distance between the matching point of 
stereo pair image is called by parallax. The 
correlation analysis method can obtain height 
data from the parallax PQ as shown in Fig.2. 
In the correlation analysis, it is very important to 
decide to the imaging angle which is represented 
the Base Height Ratio (B/H) as shown in Fig.2. 
The stereoscopic images (Nadir image and 
Forward image) for this experiment are shown in 
Fig.3. The central cross section of this terrain 
model is shown in Fig.4. 
These images are numerical height pattern using 
à computer simulation. And they are added to 
variation of the signal level (maximum 4%) 
because of imitating the natural earth’s surface. 
For this simulation, we investigate the 
correlation window size which is important 
parameter for the correlation coefficient analysis 
method. Here, the correlation window size is 
which cut partial image from each whole image for 
the purpose of comparing to correlation coefficient, 
when refer to matching points both images. The 
window sizes are shown as N1X N2 size in Fig.1. 
Figure 5 shows results of correlation analysis for 
correlation window size. If the window size 
increases, more accurate results can be obtained. 
It is evident to take a large size and a square. 
According to these results, we regard the size of 
9X9 as the most suitable size from analysis 
results and computing time. We simulate the 
following examinations with 9X 9 window size. 
  
  
  
  
  
  
  
  
  
  
  
——» M <— —— M «—— 
| 
o | E 
v] Ge b) ame | NED 
l IM — \ N , v 8 - 4 \ \ \ 
INECCOO 4I 1] 4 8h )y))) 
M NN = //] M ! $-172/]] 
^ NA A e A 
| 
  
  
  
  
  
Fig.1 The principle of correlation analysis for 
stereoscopic image. 
  
    
Nadir imaging 
  
Fig.2 Stereoscopic imaging 
  
(b) Forward image 
Fig.3 Stereoscopic image for the simulation. 
310 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
  
  
  
  
  
Fig.5