% 
have several numerical values in different channels. 
The contours of the shapes for the shape analysis are 
obtained from the absolute differences of reflectances, 
among neighbouring pixels using different channels called 
the relative gradient. The color of any shape is obtained 
from the statistical average of the reflectances of its pixels 
in different channels. 
The objectives of this investigation are: 
(1) To do unsupervised and supervised classification using 
shapes from digital images, 
(2) To evaluate the help of using this classifier in disciplines 
such as: Agriculture, Geology, Land Use, Erosion, Urban 
Planning, etc. (This work shows some examples of 
classifications but it does not show any application), 
(3) To interrelate shape and color from digital images. The 
analysis approach followed in this investigation was to 
study shape analysis capabilities. The purpose of this 
was to determine the practicality and validity of using 
shape classification. 
1 TH E RELATIVE GRADI ENT 
The definition of the relative gradient is as follow: The 
relative gradient is the absolute difference between 
neighbour pixels. Tt is possible to obtain better definition of 
the relative gradient from a digital image using more bands 
or channels (McQueen 1981); each inner pixel has eight 
neighbour pixels in different directions and orientations, 
therefore each pixels may have eight absolute diferences, 
such as relative gradients (see Fig. 1). 
It is possible to obtain less relative gradients depending 
on the selection of directions and orientations. 
Using the relative gradient, we are able to select 
different values of it; for instance, w'^at is the relative 
gradient greater than or equal to 3?. When the relative 
gradient is small there are more contours and of course 
there are more shapes, and when the relative is large there 
are less boundaries. The definition of the relative 
gradient is better when there are more bands or channels 
and I recommend using arithmetic operations of relative 
gradients among different bands; for instance, the addition 
of the numerical values of gradients from different bands 
produces a new value of gradient. 
1.1 Test site 
The shape and the color classifier was improved using a 
test site This test site is in the north of Mexico called 
"La Laguna". We employed m u 11 i s pe c tra I scanner information 
from aircraft using four channels just as the information 
from LANDSAT Satellite. 
Figure 2 shows the aerial photograph, and we can 
distinguish some properties, such as: crops, towns, roads, 
vegetation, etc. 
1. 2 Discrete representation of the relative gradient. 
Figure 3 uses a discrete representation of the relative 
gradient in two directions, x and y axis. This relative 
gradient is the addition of different gradients in different