International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV ‚ Part B-YF. Istanbul 2004 
patches and improved its quality by several pre-processing steps 
to be used by TM algorithm (Fig.3). 
  
Resampling the Template 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Y | 
Localization | 
Y ; : 
Figure 3: Desired mask 
Precise measurement 
.FA- 
o be 3.3 Image pyramid derivation 
es of v 
Is to Transformation parameters On different pyramid levels we use different representations for 
eces the fiducials. On highest levels we use the whole fiducial mark 
>s in including its surrounding(Fig.4 left) On the lower levels the 
tting 
fiducial figure (Fig.4 Right), and only on the lowest level where 
mal, Figure 1: The whole procedure of AIO final measurement is done, the same which was used for 
previous step is used (Fig. 4 Right). 
  
  
  
  
  
  
  
  
  
e by 
ct to 3.1 Extracting image patches 
both 
nple To do the job, one doesn't need to use the whole image so we 
ss of just extract the patch including fiducial mark and its 
t the surrounding (Fig.2). This helps us to reduce the amount of 
information and to increase the speed of computation. The size 
of patch is dynamically changing in correlation to the image 
size. 
ween 
inner 
letric 
igital 
al of 
le to Figure 4. Left: Pattern of fiducial mark and its surrounding. 
ation Right: Pattern of the fiducial mark figure. 
.FA- : 
dure 3.4 Detection of the orientation of the image 
isted 
In KFA-1000 photos, four of the fiducials are normally located 
in the center of each side of the photos and the fifth one is 
located in photo center as shown in Fig. 5. We employed two 
techniques of TM (Template Matching ) with combination with 
calculating moments parameters which are independent of 
image rotation to determine the tolerance rotation of (-3,3). 
in be 2 
ming 
ecise 
only. Figure 2: Extracted image patch 
1 5 3 
3.2 Resampling the template 
How to resample the template affects the success and accuracy 
of TM (Template Matching) and output data as more we 
improve the quality of template more accurate output data and 4 
0) is Template Matching algorithm will be. Here, since we knew the 
kind of camera so we resampled the template from the extracted 
  
  
  
  
  
Figure 5. Fiducial distribution on a KFA-1000 photo 
208