95 
SHAPES CLASSIFICATION ON DIGITAL IMAGES 
Ernesto Bribiesca 
Dirección de Estudios, DGG-SPP 
Cal:?. San Antonio Abad 124 
México, D . F . 
ABSTRACT 
At present, the majority of multispectral scanner data 
classifiers use statistical methods by means of the reflectances 
of each pixel (picture element). This work proposes a 
method for analysing shapes and color on digital images, the 
shapes are obtained from the absolute differences of 
reflectances among neighbour pixels using different channels 
called the relative gradient. The order of the relative 
gradient indicates the resolution of the shapes into the 
digital image. When the relative gradient is small there 
are more shapes, and when the relative gradient is large 
there are less shapes. When the shapes are found using the 
relative gradient, a procedure is developed that deduces 
from every region limited by simply connected curves a 
unique number, its shape number, independent of rotation, 
translation, size and origin. TTie order of the shape number 
indicates the precision with which that number describes 
the shape of tne curve. This procedure deduces, without 
table look-up, string matching or correlations, the shape 
number of any order for an arbitrary curve. To find out how 
close in shape two curves are, the degree of similarity 
between them is i n t r od u c ed; d iss im i I a r regions will have a 
low degree of similarity, while analogous shapes will have 
a high degree of similarity. 
The color of the shape is obtained from the statistical 
average of the reflectances of its pixels in different 
channels. 
It is possible by employing these methods to do 
automatic interpretation using shape and color into digital 
images, like people do it in pictures. 
I NT RO DUCT I O N 
Any image has many important parts for analysing, such as: 
color, shape, texture, context and relations among them. 
When people are able to recognize the differences among 
colors, shapes, textures and context, they are able to 
classify the image. 
This work proposes a method for analysing shapes and 
colors of'dig ¡tal images. Any digital image has many pixels. 
A pixel is a picture element, the smallest unit recorded by 
the sensor and there are many different sensors, such as: 
multispectral scanner, thermal scanner, etc. 
In this case, the information used was from a 
multispectral aircraft scanner employing four channels. 
In summary a digital image may have different channels 
depending on the sensor. Each channel has different pixels 
and each pixel has a numerical value, or the same pixel may