Figure 5. "hydroxyl" (left) and "iron-oxide" (right) images of the Ariab sub-scene before filtering. Block-size 53 km X 37 km. 
  
  
  
  
  
Figure 6. "Hydroxy!" (left) and "iron-oxide" (right) images of the Ariab sub-s 
potential mineralisations. Block-size is the same as in figure 5. 
3.2 GIS Overlay 
Close observation of the above composite also revealed that 
Recent unconsolidated sediments along river channels partly 
exhibit a signature similar to the gossan bodies. In order to 
  
Cl: 1 2 3 4 5 6 7 8 
1.] 17.6 8.6 4.1 13.1 1 0.4.1 0.9 1.0 0.1 
44.0 | 66.3 4.0 336 | 00.1 9.0 10.0 | 0.0 
8.2 1.2 60.9 6.1 3.5 47 l| 172 | 00 
8.2 6.2 4.4 16.0 | 1.1 3.6 4.9 0.4 
1.0 0.0 4.1 15 190.27 0.5 2.4 0.2 
10.4 | 9.4 14.7 | 1$8.| 24 | 71.2 | 341 } 00 
10.6 8.4 7.8 139 | 24 |! 10.2 | 303 | 0.0 
0.0 0.0 0.0 0.6. | 0.0 0.0 0.0 | 99.5 
  
  
  
  
  
  
  
  
OA] | A| AIR] VIN 
  
  
  
  
  
  
  
  
  
  
  
Table 4. Error matrix of the training classes used for the 
classification of the Feature-oriented PC image. The columns 
show the classified data whereas the rows indicate the reference 
data. Values are in percentage, rounded to the nearest decimal. 
Classes 1 to 8 refer to Awat-vol., basic intrusives and volcanics, 
epiclastic rocks, granodiorite, gossan, quartzdiorite, tuff/lava and 
Quaternary sediments, respectively (fig. 4). 
274 
cene after low-pass filtering. White areas indicate the corresponding 
avoid the coroboration of such undesired objects in the 
"alteration map" a GIS overlay technique has been employed. 
In this approach a ratio image of bands 5/7, 5/4, and 3/1 (in 
R, G, B, respectively) was initially computed and checked f 
the altered rocks could be discriminated. On this composite 
image the gossan bodies show a yellowish-green color in 
contrast to the country rocks and are also well distinct from 
the sediments, however not all the known mineralisations are 
mapped in the latter (Kenea, 1996). The GIS matrix-overlay 
technique is thus conducted in order to combine the 
information from these two alteration mapping methods and 
produce a map containig only alteration bodies related to the 
mineralisations. 
The above obtained two color composite images were then 
separately classified into 8 classes using supervied 
classificsation method based on a priori knowledge obtained 
from available map (figure 4). The separability of the training 
areas appears generally good and the computed error matrix 
shows negligible misclassification between the gossans and 
the unconsolidated sediments (table 4). This gave two GIS 
files with 8 classes each. The obtained classified files Were 
then examined and it has been found out that the sediments 
and the gossans have been mapped separately and the 
mixture of pixels between the two is in the order of 28% for 
the Feature-oriented PCA and 15% for the ratio images. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
is