view of the most frequent grey levels and their variance in all
spectral bands for a specific test area and may be the basis for
classification. The distribution shown in fig. 4 was taken on a
relatively large window overlapping both land and water. One ob
serves a bimodal distribution in the two infrared bands. The
lower grey level peak in each of these bands corresponds mainly
to water. A further use of the grey distribution is the possi
bility of computing a greyscale which forces an equal distribu
tion and thus enhances contrast without human manipulation.
4.4. PRINTER PICTURES
Although shapes can hardly be recognized in pictures printed out
in numerical matrix form, the numerical values are needed for cer
tain problem areas such as: reflectivity measurements, contrast
measurements, comparision between ground measurements and remotely
sensed data, calibration of thermal mapping. Fig. 5 shows a part
of our test picture printed.
4.5. TWO-DIMENSIONAL DISTRIBUTIONS
Two-dimensional distributions bring light into the dark, where
single grey level distributions do not reveal significant diffe
rences. Fig. 6 shows the distribution of MSS 4 vs.MSS 6 written
on photographic paper in the form of a 64 by 64 dot picture. The
cluster corresponding to water is of course clearly separated
from the valuesfor land, while the two different types of vege
tation are not as clearly separated.
Distributions using derived features can be still more helpful.
The distribution of the ratios (MSS 4 - MSS 7) / (MSS 4 + MSS 7)
vs, (MSS 5 - MSS 6) / (MSS 5 + MSS 6) for the window printed out
in fig. 5 is shown in fig. 7. A variety of clusters can be de
tected in this arrangement, which will not be distinguishable
in the grey vs. grey distribution.