557 
tion techniques 
al image inter 
ring blue for 
3S 7/6, and red 
enhance faults 
rver, provide a 
of different 
iced from the 
s based on the 
for lineament 
(1) lines of 
itinuity which 
¡t in images; 
of variable 
ea immediately 
aphic forms; 
6) association 
and (7) co- 
, farms, roads 
ctural and/or 
an attempt was 
ication using 
gnition. The 
he Landsat MSS 
ock types by 
eaments and by 
actural break 
beneath, the 
in alternative 
Lineaments of the South 
Central Alborz Mountains 
Monoscopic analysis of computer enhanced 
Landsat 2 high pass filtered and stretched band 7 
LEGEND 
■ — Major fault 
—|— Anticline 
Syncline 
Linear drainage pattern 
Intrusive rock 
'f' Volcano 
.... Inferred lineament 
Figure 3. Lineaments of the southcentral Alborz mountains - scale about 1:750,000. 
concept to guide mineral resource exploration and 
can be used as a complementary procedure to 
geophysical techniques for the purpose of explor 
ation. Additionally, this technique of rock discrim 
ination could be utilized in order to extend geologi 
cal mapping into unmapped and inaccessible areas 
within the region. 
It should be noted that image classification tech 
niques have not been as widely used for geologic 
applications as enhancement techniques. This is due 
to the fact that classification provides information 
on cover conditions and is affected largely by non- 
homogenity of geologic units as well as similarity of 
spectral signatures of different rock types. 
A supervised classification—maximum likelihood 
classifier—was used to identify the individual 
pixels in the scene. Training sets were delineated 
on the computer's color display screen with polygon 
programs. The selection of training sets was aided 
by field geology information and topographic maps. 
Data on the sample means, and variance-covariance 
matrices were derived from training set statistics. 
Several statistical programs, including training set 
check and training set divergence programs, were used 
to evaluate spectral separability by creating confus 
ion matrices and computing transformed divergence for 
each pair of training sets. The transformed diver 
gence analysis procedure is described by Haack (1984) 
as follows: 
Transformed divergence, which is calculated from 
the means and covariance matrices of each spectral 
class or training site, is a measure of the stat 
istical distance between class or site pairs of 
interest and provides information on their "separ 
ability". This separability is an indirect 
estimate of the likelihood of correct classifi 
cation between groups of different band combina 
tions. Such an estimate provides information 
usually obtained by the time consuming and 
expensive process of actual classification and 
accuracy evaluations. Transformed divergence can