din 
ori 
icated 
hat a 
From 
je se- 
dency 
for the 
> the 
consid- 
This 
a the 
annels 
ctions 
t is 
a modi- 
(T) 
     
This quantity considers only the sign of the residuals, how- 
ever not their possible dependency on the position or type 
of the scanned object. The establishment of a corresponding 
functional model would have requested an unreasonably high 
expenditure. In addition, it seems to be very questionable 
whether or not the different topographic objects, which for 
the linear dependency does not apply, could have been de- 
scribed with satisfying accuracy by a single function. In- 
stead of this, the residuals inserted in the test mentioned 
above produce a new image with the density values G1, G2, 
and G3 obtained from the following decision rule: 
if 
tv, > + b etimol» then v, results in G1 = 255 
(A full blackening) 
- NL s 1; € by Im |, then v, results in G2 = 128 
(A medium density) (8) 
ev t sols then v. results in 05 - © 
(^ white), 
whereby t was set to 1.0; 0.5 and 0.1. (With ts 3.30 for 
a statistical certainty of 99.9% only clouds were represent- 
ed.) Figures 2 - 7 show the results. 
The decreasing t,-values effect that the indifferent areal 
elements primarily with [e = 1.0 in medium density become 
cither white or black pixels, thus resulting in noisy or 
linearly or arealike structurized images. The selected exam- 
ple demonstrates that not all four channel combinations of 
the areas 1 and 2 produce noisy images, as could have been 
expected from Table 2, but only the combinations of the chan- 
nels 4 and 5. This yields result 2: 
Two spectral data sets depend linearly on each other only, 
if the selected functional model fits, whereby mo attains 
the magnitude of the digitizing errors conditioned by the 
System. Due to the fact that the pixels computed from the 
residuals according to equation (8) are independent of their