a line 
can be 
id that 
ioned) 
ching. 
tching 
af de- 
targets 
ethods 
s mat- 
1987, 
| gives 
riance 
atively 
of the 
à quite 
to re: 
tching 
ation 
dies of 
digital 
TE cite 
asis of 
‘char 
18 was 
8,7.5, 
values 
r each 
(grey 
which 
y. The 
r each 
| (100 
0f the 
r up to 
The 
(me ) 
ated in 
ter of 
m,are 
es de- 
he left 
ratios 
elsize 
taget 
ilar to 
andom 
uld be 
duced 
ulated 
ation . 
+10% , 
ratios 
1. The 
ization 
better 
| size) 
; equal 
>rihan 
1:10). 
',1989. 
me 
our (pixels) 
| 27 
oO l2r 
0.107 
17 
0.087 
6.8 
0.067 
0.04} 
0.02F 
  
  
  
Fig2 The relation between the precision of a target location and a 
quantization level ; 
  
4 Me 
ixels 
ou | (p ) 
0o,l2 Fr 
0.10% 
0.087 
0.067 
0,04 F 
3 
0.02} A 
15 
4 i À i l L 1 1 26 K 
  
Fig3 The relation between the precision of a target location and 
ratios of signal to noise. 
In order to test the influence of local noise on the 
precision of a target pointing the following ex- 
periment was perfomed. The initial datas are: win- 
dow sizes - 42*42 pixels, pixel size - 2.8um, circular 
target size - 100pm. quantization level - 256 (29). 
random noise - 10% . The local noise was simulated 
as square (in the left hand corner of window in figure 
4) which have the same grey values af pixels like a 
target. The sizes of this local noise were changed. 
  
  
  
  
a 8 
Fig4 The circular target with the local noise. 
The figure 4 shows (schematicly) minimum and 
maximum sizes of local noise which where ap: 
proximately 1:40 (fig.4a) and 1:4 (fig.4b) of circle 
length. 
    
   
  
   
    
   
  
  
  
  
  
  
     
     
   
  
     
  
   
    
   
    
   
   
   
    
  
    
    
    
    
   
     
   
   
    
    
     
      
    
  
   
    
   
     
  
   
  
    
      
   
    
A Me 
(pixels) 
0.057 
0.041 
0.037 
0.02} 
0.017 
  
L 
5 10 20 30 40 
Fig5 The relation between precision of target location and local 
noise. 
The figure 5 demonstrates the variations in pre- 
cisions of pointing due to local noise. Here L is the 
ratio of circle length to arc length which is covered 
by noise. At the same time (as a comparision) the 
precision of target pointing without local noise was 
0.005 pixel. 
The local noise doesn’t influence the precision of a 
target location when DO. This means that robust 
algatithm (2,3) is rather effective to suppress noises. 
The similar investigations where caried out for cross 
and square targets. 
Table 1 illustrates only some results of cross and 
square targets location precision (in pixeis) as 
compared with a circular target, T he initial datas for 
tests presented here ars as foliows: a quantisatioun 
level — 32 12%) a random noise - 10% , a ratio to 
target size/ pixel size - 8,15,36. The relationship 
betwesn the pointing precision and the quantization 
level, random noise for a cross and a square targets 
is in many ways similar to that of a circular ane. 
Table 1. Precision of target pointing (pixels) 
Kztarg.size/ pix.size 
Target lemen miners tm ce E c d EE 
8 15 36 
circle 0.03 0.02 0.01 
cross 0.27 0.08 0.03 
square 0.54 0.31 0.20 
Table 2. Precision of target matching (pixels) 
K=targ.size/ pix.size 
Tarvet en 
8 15 36 
circle. 0.03 0,02... 0.01 
cross 0.08 0.04 0.02 
square 0.17 0,15 0.13