nage coordi- 
5. 
  
|ter design 
-band scan- 
e frequency 
same effect 
.6. Instead 
behaviour. 
filtered 
equency do- 
um) and 
sed. 
(6) 
imagery 
ere is a 
early ful- 
rding to the 
on H(u,v) 
k to avoid 
tion in the 
es 
z" po N : 
4 yam sn = ee 
ee de s m SSmo a - = : — 
  
frequently domain. The filtered image is obtained by inverse Fourier-transform- 
ing the product of F(u,v) and H(u,v) (SFB 149, 1980). The result is shown in fig.8. 
The high frequent scan line noise disappears nearly completely, but the filter is 
relatively raw, so that interfering effects are to be séen. So there must exist 
another way for optimal filtering. 
Ampl i tudenquadrat * : I/Io 
  
  
  
1... I 
7 rr 9.5 : T f/fu 
  
  
  
0.0 0:5 1.0 f/fu 
0.75 4 
Fig. 6: Power spectrum of a column 
  
Fig. 7: (on the right): Powerspectrum Kos 
channel 11 (upper) | 
channel 9 (middle) ET. 
Ratio 11/9 (lower) ; ij" 
  
  
  
Scanner-Regression-Filter 
For the design of a filter, which is only valid for scan line noise, one has to con- 
sider that this noise is column independent. The distorted image signal s' is formed 
by the original signal s and additional noise am according to 
s' (n,m) = s(n,m) + Ps (n) (7) 
This can be justified by looking at fig.9, which shows a profile of a scan line 
compared to a profile of an image column over water areas. 
It can be seen that the variances in column-direction are by two orders of magnitude 
higher than in scan-line direction. 
Using (7) a scanner regression filter (SRF) is developed depending only on line 
number and magnitude of signal s' for optimal signal restoration. 
S' (n,m) + SRF (n,s) = s(n,m) (8) 
2139