distinctive white and yellow colors.
In addition, an ISI System 13o was used for producing analog
color equidensity images whenever grey tones had to be compared
over larger distances. The images produced on the color monitor
| were photographed by a HASSELBLAD camera and then superimposed
| on the original visual interpretation by means of a BAUSCH &
LOMB ZTS-4H Zoom Transfer Scope.
Finally, digital classification procedures were used for
selected areas where interpretation was difficult (fig.3 and 4).
The digital image processing had two main objectives: First,
to produce image enhancement of areas where the contrast was
so low as to make interpretation difficult, and secondly, to
classify areas where spectral reflectances in a single band
did not show enough differences for an unequivocal visual
identification.
For the supervised computer classification, test areas were
selected in the images of which reliable ground truth was
available. A minimum distance classifier was compared with a |
hyperbox classification in order to identify those areas
| where "clean" spectral signatures occured. A. common difficulty
| arose from the extensive cover of wind blown sand on the
exposed bedrock so that large areas produced mixed signatures
which were difficult to eliminate.
In addition, there is a high correlation between the four
Spectral bands of LANDSAT (r = 0.98 - 0.99). The problem is
aggravated by the widespread occurence of desert patina or
weathering crusts formed on the surfaces of the exposed rocks
which lead to a masking effect that equalizes the spectral
responses.
Thus the digital classification on the base of spectral signa-
tures can, at the present time, only give additional informa-
tion to the image interpreter. Further developments in the
use of image texture analysis (BÜRGER 1977) will hopefully