34 
A detailed comparison yields small differencies 
between both images, e.g. at the shadowed parts of 
the central Basalt complex. The feature space in 
Fig. 5 c indicates an unfavourable arrangement of 
some classes in comparison to their arrangement 
in the original - or optimized data (Fig. 5a,b). 
This could lead to more confusion in the classifi 
cation. 
The application of the IHS-components for the 
classification did not result in an improvement of 
classification. 
3 COMPARATIVE ESTIMATION 
In general, both methods presented, cannot be seen 
as different techniques which lead to comparable 
results. 
As shown above, there are several areas of overlap 
during processing. For instance, the choice of 
training areas needed for supervised classification 
is carried out visually by enhanced images. On the 
other hand, various statistical data are used to 
select processing parameters for image optimization. 
It can be stated that for applications where 
structural information is of high importance 
(geology), optimized image products are prefered 
For data acquisition of various surface signatures 
over wide areas (landuse), the use of classification 
methods seems to be more advantageous, particularly 
as relative proportions of single categories are 
determined simultaneously. The essential advantages 
of both methods can be summarised as follows: 
- The unsupervised classification is an objective 
method for separation of different signatures 
on a statistical basis. This is also true of su 
pervised classification, even though the choice 
of training areas is of subjective character. 
- A highly selective reduction of information can 
be made for any particular surface category, in con 
trast to image processing algorithms (PC's or 
ratios), where reduction depends on choice of bands 
and processing parameters. 
- Any color can be assigned to the resulting cate 
gories so that neighbouring classes are well 
differentiated. 
- Percentages of all classes are determined simul 
taneously. 
- The principal advantage of image optimization 
products lies in the fact that relief information 
is preserved. Such products are a portrayal of 
the earth's surface, which can at least be evalu 
ated and interpreted in photogeological terms. 
- Besides the possibilities for planning and 
execution of field work (classification results 
are not usable due to the absence of reference 
points), several strategies for structural ana 
lyses become feasible. 
- In contrast to classification results, enhanced 
products do not present a final result. Conse 
quently, the subjective character of data evalu 
ation can be left in the hands of the interpre 
ters. This can be seen as an significant advan 
tage. For drawing conclusions on, for instance, 
genetic events, a simultaneous evaluation of 
structural and spectral features is indispensable. 
- Furthermore, the recognition and localization of 
unknown spectraldiagnostic enoinalies is only 
obtainable by choice of suitable bands, and not 
by given training areas. 
- The comparative assessment has shown, that ad 
vantages of one method are not necessarily a dis 
advantage in using the other one. From a compu 
tational point of view there is little differen 
ce, and the choice of methods depends on the 
required objectives. The principal advantages of 
both methods can be summed up as: 
- An objective computer supported classification. 
- Preservation of relief information during the 
optimization process. 
Tab. 2 Statistic of classified training aeras 
Version a: original data 1/4/7 
Version b: optimized data 1/4/7 
Version c: IFIS-components 
No. name 
i 
ver- 
I 
i 
2 I 
i 
3 
i 
4 ! 
5 
6 
7 
8 
9 
1°i 
11 
12 
13 
14 
sum 
sion ! 
i 
3ga 
sba 
sbb 
sbc ;tix 
ea 
jra 
grb 
xya 
xyb j 
tbra 
il 
tbx 
gp 
1 gga: 
I 
a 
178 
1 
179 
Granite 
b 
178 
1 
179 
Granodiorite 
C 
178 
1 
179 
2 sba: 
a 
204 
2 
206 
Schistose 
b 
202 
1 
3 
206 
complex 
C 
204 
2 
206 
3 sbb: 
a 
71 
1 
72 
Schistose 
b 
71 
1 
72 
complex 
C 
71 
1 
J- 
72 
4 sbc: 
a 
1 
307 
308 
Schistose 
b 
j 1 
307 
I 
308 
complex 
C 
i i 
i 307 
I 
308 
5 tix: 
a 
69 
69 
Laterite 
*> 
69 
69 
(Kaolinite) 
c 
69 
69 
6 vea: 
a 
39 
39 
Vege- 
b 
39 
39 
tation 
C 
39 
39 
7 gra: 
a 
160 
160 
b 
160 
160 
Granite 
C 
160 
160 
8 grb: 
a 
68 
68 
b 
68 
68 
Granite 
C 
68 
68 
9 xya 
a 
63 
1 
; 1 
65 
b 
63 
1 
1 1 
65 
? 
C 
63 
2 
65 
10 xyb 
a 
1 
77 
78 
b 
1 
77 
78 
? 
C 
1 
I 
77 
78 
11 tbra: 
a 
1 
60 
61 
b 
1 
60 
61 
Ryolithe 
C 
1 
60 
61 
12 til: 
a 
22 
22 
Laterite 
b 
22 
22 
(Limonite] 
C 
22 
22 
13 tbx: 
a 
1 
105 
106 
b 
1 
105 
106 
Basalt 
C 
1 
105 
106 
14 gp: 
a 
322 
329 
Alkali 
- 
b 
1 
¡323 
329 
granite 
C 
1 
'322 
¡329 
Fig. 7 shows classification results in the Karlsruhe 
area. Fig. 8 shows an example of combining the 
advantages of image optimization and classification. 
This combination has been achieved using the IHS 
approach, were intensity (I) represents relief and 
Flue and Saturation (H,S) are derived from classi 
fication analysis. 
This can be seen as a suitable strategy for 
future developments.