Fig. 4. The original and the normal image 
  
  
  
  
  
ee" e@ | 7 C 
PV, n 
e 
     
° @ OC 
€ * 92s! |« 
Fig. 5. The enlarged grey value, binary and contour image of one target 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
cluster 
size group 1 à small squares and 2 small triangles 
size group 2 2 small crosses and 2 small circles 
size group 3 2 large squares and 2 large triangles 
size group 4 2 large crosses and 2 large circles 
  
  
  
  
Table 3. Division of the targets in size groups 
For each size group, the shapes of the targets were determined by an 
analysis of the contours. Counting the frequencies of features like "sharp 
angle" and "concave angle", two likelihoods were derived for each target: one 
for the target being of type one (e.g. a small square) and one for the target 
being of type two (e.g. a small triangle). The relations between the frequen- 
cies of the features and the likelihoods were taken from a look-up table, 
which has been made manually after examining a few test images. 
Having four targets within one size group, there remain six possible 
combinations of classifications (knowing that there are two targets of type 
152