774 
N 
G. 
i, 
W, 
The Class covariance matric Wç which is a measure of Class dispersion within 
the training area is defined by: 
where diagonal components are variances of distributions in single spectral 
bands of Class C and the off-diagonal components are covariances of 
distributions in combinations of two spectral bands. 
The mean spectral vector Gç and covariance matric W c define the 
multivariate gaussian probability density function of Class C (Landgrebe et 
al., 1968; Scherk, 1974): 
M 
(G, 
i J) 
(2tt) N/2 |W. 
exp [- i (G. 
G / V 1 ( G i,j 
G c )] 
The 
class 
class 
to th 
class 
undei 
"unkr 
thema 
singl 
based 
were 
digit 
compu 
sepai 
speci 
large 
test 
close 
image 
mediu 
film 
this 
satel 
by t 
succe 
DICOM 
NW is probability of occurrence of spectral vector G^ • in 
iJ Class C 
The m 
class 
from 
chang 
great 
where