Byung-Uk Park 
  
3 SUBIMAGES OF FIDUCIAL MARKS 
The aerial photo was taken by Wild RC30 Camera. The nominal focal length is 152.85 mm and scanned aeria] 
image scale is 1:5,000. The diapositives were scanned at a resolution of 600 dpi by VEXCEL corporation's scanner, | 
was scanned with a pixel size of 42 un, 8 bits per pixel, yielding a ground resolution 21.2 cm. The total scanned image 
size is 5481 pixel x 5378 pixel. 
For the locating of four fiducial mark using Hough transform, subimages containing fiducial marks are 
manually extracted. Size and position of each extracted image patch are described in Table 1. Each extracted subimage 
of the fiducial marks is shown in Figure 3. 
  
  
  
  
  
  
  
  
  
  
  
  
Fig No. | Size of Image patch | Image coordinate of patch image in full scene 
Fig 3a) | 55 pixel x 52 pixel NA re Lill 
Fig 3(b) | 56 pixel x 55 pixel ak Drm. CT Om 
Fig 3(c) | 55 pixel x 54 pixel RN 2 = 
Fig (a) | 53 pixel 51 pixel | REP — 6222-5506) 
  
  
  
Table 1. Size and position of image patch 
     
(a) | (c) 
Figure 3. Subimages of fiducial marks. (a) upper left corner; (b) 
4 EXPERIMENTS AND RESULT 
We assumed to know the position of the fiducial marks approximately according to the camera type and we 
extracted manually subimages around fiducial marks as shown in Figure 3. The extracted subimages around the fiducial 
mark are transformed to binary images for the Hough transform. Simple thresholding technique such as predetermined 
threshold value is problematic since the threshold value can be varied locally as shown in Figure 3. We implemented 
local dynamic thresholding algorithm in this research. 
The local dynamic thresholding method obtains the local threshold values based on the relative local intensity 
distribution compared to global thresholding that gets thresholding values based on the whole intensity value. Detail 
description of the algorithm is presented in Sohn (1996) and the size of local region to estimate bimodal approximation 
was 8x8. The algorithm automatically converts input images into two classes (foreground and background). The results 
of the two class images of the input subimages are shown in Figure 4. 
Rs] pe] [59] [d] 
(a) (b) (c) (d) 
Figure 4. Binary images of Figure 3 using local dynamic thresholding. 
  
694 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
used © 
binary 
succes: 
transfo 
fiducia 
central 
central 
fiducial 
Table 4