re the 
rmine 
pixel 
S on 
ive to 
> filter 
ontal, 
ds of 
holds 
>nded 
S are 
ts are 
least 
-point 
given 
ection 
hip of 
major 
(see 
ediate 
steps 
on of 
| and 
which 
ghtest 
pixels 
strong 
  
  
For an exemplary detection of tie-point candidates, all 
these preprocessing steps have been applied to the test 
image. In a next step the geometrical ratio operator was 
applied in order to detect edges in the four main 
directions with a 5 x 5 filter kernel. The respective results 
are shown Figure 3 (a-d). These images are further input 
for the another ratio operator, which determines 
likelihoods for the individual pixels to be corner points 
along an edge. The result of the operator is shown in 
Figure 3 (e), where the likelihood increases with the 
darkness of the pixels. 
Finally Figure 4 shows the tie-point detection result, which 
has been generated for the selected image chip. The 
detector allows the definition of a number of „best“ tie- 
points to be found in the selected area. In the present 
case the detection of 15 and 5 tie-points, respectively, 
was envisaged (see Figure 3, left and right). A minimum 
distance of 40 pixels between individual candidates was 
further specified. 
  
  
  
  
  
  
  
  
  
  
  
Figure 3: Results of GRO, applied in 
horizontal (a), vertical (b) and diagonal 
direction (c and d), and of ratio 
edge/corner point detector (e). 
  
  
  
317 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996