RULE-BASED SYSTEM FOR UPDATING SPATIAL DATA-BASE 
Basheer Haj- Yehia, Ammatzia Peled 
University of Haifa, Department of Geography, Haifa, 31905 Israel 
basheer@geo.haifa.ac.il; peled@geo.haifa.ac.il 
Commission II, IC WG IUTV 
KEY WORDS: GIS, Mapping, Change Detection, Classification, Recognition, Updating, Knowledge Base. 
ABSTRACT: 
Spatial Information revision and updating is the main concern and production effort of maintaining the ever-growing GIS systems 
and spatial data bases. Developing easily effected automatic updating methods of spatial information becomes the key to the 
successful maintenance of the large GIS data bases established by many mapping agencies all over the world. The objective of this 
research was to develop a rule based system for updating a spatial database. Taken into consideration were rules, such as: (a) The 
radiometric and textural parameters; (b) The geometric parameters of the objects, such as area, perimeter, compactness, elongation, 
ete; and (c) The topological relationships between objects of the same type group and from different layers, as well. This paper 
presents the algorithms and rules which were implemented for updating the Israeli National GIS spatial database. Also, discussed are 
the experiments implemented over the Haifa region test site. 
1. INTRODUCTION 
Updating of spatial databases can be carried out by complete, 
new and updated mapping, which replaces the existing old 
information. Alternatively, one may gather spatial details which 
are not found in the existing database. Their addition to this 
same database may constitute the major part of the updating 
process. The advantage of the last approach is the small amount 
of objects which require treatment. In this approach, databases 
updating processes are composed by three stages: (a) Change 
Detection - Finding objects and regions of change; (b) Change 
Recognition and Identification - Determination of the character 
and type of change; (c) Revision - Introducing the identified 
changes to the spatial database, while preserving the topology 
and the structure of the database. 
Most of the traditional change detection and identification 
methods are based on the radiometric information from satellite 
or airborne remotely sensed data [Mouat, et al, 1993; Peled, 
1993; Muchoney & Haack, 1994; Jha & Unni, 1994]. These 
methods are based on a comparison between the grey level 
values on two images from different dates. The common 
method of digital image comparison is image differencing. In 
this method, grey level values on two images are subtracted for 
each pixel. If the absolute value of the subtraction is greater 
than a specified threshold then the pixel is defined as a 
"changed" pixel, otherwise it is marked as "no changed" pixel. 
The major problem of these methods is the accuracy level of 
separating regions of change/no-change and identifying the type 
of change. In most studies, the thresholds are set empirically. To 
improve these methods, the analysts are looking for more 
automation. On the other hand, the automation processes, suffer 
from complex and uncertainty of the objects recognition. In the 
last years, most of the studies of change identification use the 
post-classification approach. In this approach two images of 
different epochs are classified and then, a pixel by pixel 
comparison is implemented. An example of these studies is the 
work of Muchoney & Haack [1994]. The major effort of these 
studies was to improve the classification method. Other research 
studies developed fuzzy logic algorithms to improve the 
classification methods [Bellacicco, 1996; Warner & Shank, 
1997, Metternicht, 1999]. The problem of these works is the 
fact that they were based on single pixels classification, without 
taking into consideration entire objects. In recent years more 
and more identification methods were developed by using 
segmentation techniques to extract separate objects from the 
images [Tilton, J.C., 1998]. In this study [Tilton, 1998], the 
image is segmented using only radiometric criteria by using the 
Euclidean similarity distance. In general, most of the studies are 
focused only on a specific issue, such as improving the 
classification method, identifying thresholds or detecting 
changes for forest monitoring, rather than updating GIS 
databases. 
The purpose of this 'rule-based' research study was the 
development of models and methods for updating the 
Geographical Information System (GIS), combining 
information from different sources, efficiently and 
automatically, as possible. The research focused, mainly, in the 
Change Identification stage. To achieve this objective, a rule 
based system was developed and examined. This system was 
built to integrate and fuse data from various sources. Taken into 
consideration were rules, such as: (a) The radiometric and 
textural parameters; (b) The geometric parameters of the 
objects, such as area, perimeter, compactness, elongation, etc; 
and (c) The topological relationships between objects of the 
same type group (class) and from different layers, as well. 
2. METHODOLOGY 
In this research, updating the spatial database is carried out 
based on the “detecting and identifying changes” approach. 
According to this approach the process of change detection was 
implemented in four major steps: (a) detecting regions which 
were changed; (b) calibration parameters and thresholds of the 
rule-based system; (c) segmentation of the regions of change to 
separated objects; and (d) identifying the type (class) for each 
changed object using the rule-based system. 
A rule base system was developed for change identification.” 
This system is based on four different types of parameters: (a) 
Spectral parameters; (b) Geometric parameters, such as area, 
perimeter, compactness, moments of inertia, etc.; (c) Textural 
parameters, such as contrast and homogeneity; and (d) 
Topological and spatial relations between the objects. 
According to these parameters, a set of rules was defined for 
each type (code) of objects. The rules were developed to 
describe each type (code), uniquely. For instance, a building is 
408 
Int« 
des 
sha 
by: 
oth 
The 
hie 
alg 
obj 
det 
sin, 
lab 
seg 
dif 
rad 
dat 
usi 
In 
spa 
gel 
anc 
for 
anc 
adc 
wa 
the 
in | 
anc 
err 
sys 
for