649 
GIS-Guided Road Extraction 
form Satellite Imagery * 
J. Van Cleynenbreugel F. Fierens P. Suetens A. Oosterlinck 
ESAT-MI2, Katholieke Universiteit Leuven 
Kardinaal Mercierlaan 94, B-3030 Heverlee, Belgium 
email : vanclevn@esat.kuleuven.ac.be 
Abstract 
We present a (semi-)automatic system for road delineation on high resolution 
satellite images that has capabilities to integrate different GIS-related knowledge 
sources. The system is developed in an object-oriented image understanding envi 
ronment based on an existing knowledge-engineering tool (KEE). In this way, GIS 
data are easily represented and expertise can be added flexibly. The feasibility of the 
concept has been tested on three practical case studies using SPOT and LANDSAT 
TM data. 
1 Introduction 
Human analysts rely on expertise in combining external data (such as topographic maps 
and landcover classifications), to solve a spatial image recognition problem like the ex 
traction of roads and linear networks from remotely sensed imagery. Although such data 
are now typically stored in a GIS environment, existing approaches to automatic road 
extraction from satellite images have been hardly based on any knowledge related to GIS- 
datasets. The following data sources and accompanying knowledge are important to human 
experts to delineate road networks : global landcover classifications, existing roadmaps and 
hydrographic maps and terrain (elevation) models. Depending on the complexity of the 
task in terms of road-density, road-complexity and image quality, experts can decide what 
sources must be involved. In this paper, we present three case studies employing different 
knowledge sources. 
The first case study principally deals with landcover-related knowledge. Delineating 
forest paths in flat terrain is a typical example. A generic model describing the appearance 
"The following text presents research results of the Belgian National incentive-program for fundamental 
research in Artificial Intelligence, initiated by the Belgian State - Prime Minister’s Office - Science Policy 
Programming. The scientific responsability is assumed by its authors.