795 
* + * * + 
******** 
<*- 
<NJ 
be described by a distance along an axis corresponding to that 
band, a point may be located in space which is the integration 
of any possible combination of brightness levels in the different 
bands, and which is unique to that particular combination. 
Thus, we have a quantitative description of the color of any 
cell in the scene, which can then be used in several ways to 
classify it. (This concept is simply a mathematical framework 
for describing what many are more familiar with as the "spectral 
signature".) 
In supervised classification, two techniques are commonly 
employed to discriminate cover types based on this four-space 
construct, one involving the angle between lines drawn from the 
origin to the points in space representing the colors of the 
cells, and the other calculating the distance between the 
points. In visual terms, the angle criterion is based upon the 
hue of the cell, while the distance method relies upon the inte 
gration of hue and total brightness components. In practice, 
the user selects areas of known class (training areas), and 
statistics are computed for each class, which can then be used 
as standards against which unknown cells are compared. 
APPLICATIONS 
The package of basic programs described above was employed 
for the study of numerous sites in the eastern United States, 
by three research groups, and in several disciplines. The 
author’s investigations included a study of the merits of re 
latively simple contrast enhancement procedures, combined with 
human interpretation, to map the uses of land in an urban area 
with which many are familiar (Figure 3). The figure illustrates 
one component of the computer’s contribution to a final land use 
map: an enhanced image of the study area for a particular 
band, which may then be interpreted for various features (in 
this case, vegetation). In combination with similar images 
from other bands which illustrate the urban structure more 
clearly (note that Figure 3 represents all urban uses as a 
single class), this gray-scale picture will yield a land use map. 
A second, more extensive, application was undertaken on a 
1500 square mile area in Eastern Pennsylvania, comprised of 
Lehigh, Northampton and Berks counties, as part of a Regional 
Environmental Assessment Procedure (REAP) developed for the 
Commonwealth of Pennsylvania, under the auspices of the Office 
of Planning and Research, Department of Environmental Resources 
(Dhar, et al., 1974). In this instance, the techniques of con 
trast enhancement and the joint frequency distribution were used 
to create "working maps" of selected areas of the site, from 
which training areas were selected for the supervised classifi-