viations between the manual measurements and the matchings are marked 
with "+". 
The relation between win- 
dow size and standard de- 
viation of the parallax 
for the sub set Rock(44). 
No geometric parameters 
are used. The optimum 
number of observations 
was 615 pixels. corre- 
sponding to the optimum 
window size 25X25 pixels. 
The estimated variance 
components _were 65, 36 
and 15 um for the me- 
thod, the noise and the 
height dependent compo- 
nents, respectively 
The relation between win- 
dow size and standard de- 
viation of the parallax 
for the sub set High(12). 
No geometric parameters 
are used. The optimum 
number of observations 
was 917 pixels.  corre- 
sponding to the optimum 
window size 30X30 pixels. 
The estimated variance 
components yere 19, 160 
and 30 um for the me- 
thod, the noise and the 
height dependent  compo- 
nents, respectively 
The relation between win- 
dow size and standard de- 
viation of the parallax 
for the sub set Rock(30). 
Affine geometric parame- 
ters are used. The opti- 
mum number of observa- 
tions was 837 pixels. 
corresponding to the 
optimum window size 29X29 
pixels. The estimated 
variance componepts were 
58, 17 and.^ um - for the 
method, the noise and the 
height dependent  compo- 
nents, respectively 
lum) 
The curves result from the variance component regressions. 
  
  
15 
  
T 
30 40 50 
  
  
212714 20 
Window Size 
(pixels) 
g 
(um) 
A 
" + 
+ + 
1 1 I i 1 T Wu 
0 12 34.420 30 40 50 
Window Size 
(pixels) 
a 
(um) 
+ 
+ 
T T T T "T T^ 
0 12 16 20 30 40 50 
Window Sıze 
(pıxels) 
It seems as it is possible to find an window size with minimum r.m.s. 
with a satisfying accuracy, but the results will of course be restric-