which is used in the following way: 
a) having examined a figure (cluster of pixels) from the 
image, the area and the perimeter can be determined; 
b) on the basis of the area and perimeter number the indi 
vidual is then attributed to one of the classes. For example, 
for an element the area is55 and it therefore belongs to the 
first class: the individual can be a class B olive tree or an 
unidentified object. 
Thus the perimeter is 38 and hence belongs to the class 3 
as far as the perimeter is concerned; 
c) the operator weights up the two pieces of information 
and identifies the individual as a non-olive tree. 
When there are more than two characteristics to be combined 
various operators are studied and combined in a function cal- 
led the "multivalue logical tree" (see later, . Table 1). 
Operators of this kind are used for evaluating information 
relative to one single object, but surveyed in different bands. 
Each one of the surveys obtained in the different bands 
constitutes a scene in itself. 
It sometimes happens that the same object appears with dif 
ferent characteristics in different bands; sometimes an object 
may not. appear, Or 4t may be broken down into subclusters. 
Hence a wider operator is called for, one which can discover 
which sets of points, in the different images, should belong 
to one particular tree (operator of Inclusion). 
2. Available data. 
For this job, which involved the combined use of spectral 
recognition and recognition of form, a false colour image was 
used, a Kodak 2443 film with a Wratten 12 filter. 
The image -transparent- was then scanned with an Optronics 
Photomation 1700 instrument with a 50 micron spot size (corre 
sponding to a 50 x 50 centimetre pixel on the ground, the ima 
ge being in scale 1:10.000) in the three basic colours blue, 
green and red. 
The analog scan signal was then sampled every 50 micron and 
digitized choosing 256 levels on a density dynamic of 3, and re 
corded on magnetic tape. 
This approach offers, contrary to what happens with a multi 
spectral scanner, a low number of available bands, but an extre 
mely high geometric solution (50 x 50 centimeters on the ground 
from a flying height of 1500 metres corresponding to an "Istan 
taneous Field of View" on the scanner equal to O.3 millirad. 
2.1 The performed exploration. 
We choose four basic parameters: perimeter, area, ratio perime 
ter/area, ratio number of vertices/perimeter. 
Seven classes for each parameter were selected from the ana 
lysis of data.