ET M cs 
To have truly representative statistical values for the actual brightness situation, 
U. and C. must be a function of scan angle « 
e.g. with following function: 
2 
*« 
U. = U, + U «X + U, 
io i i2 
i 1 
To fulfill this requirement in an computer program polynomial coefficients for means and 
covariance matrices for each class and wavelength must be supplied. These coefficients 
can be processed in two different ways. In the first case values of these polynomial functions 
in arbitrary scan angle rates are calculated and stored in tables before classification. During 
classification the appropriate statistical values can be selected from these tables as function 
of scan angle, With this solution more storage is needed but computer time increases only 
slightly. For example 100 different mean values and 30 different covariance matrices per 
class should be stored in this case. In the second case values of these functions are 
calculated as required during classification. This solution needs about twice as much 
t 
computer time. The first solution was realized in the program. 
190 
3.2 D 
The po 
possibi 
set usii 
directi 
using t 
be regi 
Furthei 
(perper 
Mean » 
Trendc 
polyno 
determ 
3.3 R 
To test 
chaptei 
statistic 
covaric 
deciduc 
Table 1