105
3.4 Methods and techniques for data
processing
In our Centre for Computer Graphics and
Mapping we are using different GIS software
packages. In this respect we are investigating
all the different functionalities needed for good
GIS systems.
In respect to the data processing our re
search is aimed at
- knowledge based systems;
- computer vision;
- neural networks;
- object oriented grid technology
3.4.1 Knowledge based systems
In computer science there are two develop
ments who are now combined and are
strengthening each other. These are the
database technology and the knowledge
systems. The combination is calied a know
ledge base. These systems, also called expert
systems, are developed to assist an expert in
his work or even to replace him. This is
important for GIS/LIS, because this is an
expanding field of science where the number
of experts is rather low. The application of
expert systems in spatial information was up
to now oriented on the creation of cartogra
phic products by poor educated map makers.
With the introduction of expert systems in this
area, two problems arise:
- the knowledge and skills concerning map
making is net well described. If two carto
graphers are producing a map from a
special area, than the results are completely
different. The procedures necessary for
expert systems are probably not well de
fined;
- geographers, cartographers and land sur
veyors are using expert systems mainly to
produce maps. If non-experts, without any
affinity to maps, are using expert systems
than they try to solve the problem with a
limited use of graphic components. These
experiences I have nowadays in my own
research centre. Mathematicians and com
puter specialists are busy with the introduc
tion of expert systems in spatial information.
The use of maps and graphic symbols is
rather limited.
3.4.2 Computer vision
Computer vision is a branch of artificial intel
ligence, dealing with the construction of
explicit, meaningful descriptions of physical
objects from images. Computer vision systems
are the artificial vision systems that are ca
pable of understanding indoor or outdoor
scenes at a human level. The final goal of a
computer system should be able to look at an
image and describe the scene depicted in
words.
Computer vision is a relatively new research
area and has received a growing interest from
different disciplines. It is a kind of difficult and
challenging artificial intelligence to achieve
because of the following reasons:
1. The world is three-dimensional, and the
images from which a description must be
formed are only two-dimensional projec
tions.
2. Each pixel of an image represents the
interaction of many processes, and it is
difficult to separate these different influen
ces such as illumination, reflective proper
ties and geometric distortions, etc.
3. The volume of data in a good image is
very large.
4. In order to interpret an image intelligently,
much knowledge is needed about the
objects that may appear in the scene.
Generally, research on computer vision is
constructed as a three-level integrated raster-
vector processing system. At the first level,
grey values (raster data) are processed (low
level) from which structures (features) can be
extracted and manipulated as symbolic des
criptors (mid-level). At the highest level, know
ledge based information often coupled with
spatio-temporal models gives a predictive
description.
In the field of surveying, mapping and GIS/
LIS, computer vision can at least find its
applications on the following aspects:
- integration of remote sensing with GIS/LIS;
- the automated surveying system, for ex
ample, the CASSPAR system;
- automated feature recognition from scanned
maps;
- automated photo-interpretation of aerial
photographs and remote sensing image.
3.4.3 Neural networks
Neural network belongs to one of the promi
nent research topics of the scientific com
munity in the 1980’s. Artificial neural net-
models have been studied in the hope of
achieving ‘human-like performance in the fields
of speech and image recognition. These
