ROAD DETECTION FROM HIGH AND LOW RESOLUTION SATELLITE IMAGES 
R.Gecen 3 , G.Sarp b 
“"Department of Geodetic and Geographic Information Technologies, Middle East Technical University, 06531, Ankara 
Turkey - rgecen@metu.edu.tr 
b Department of Geodetic and Geographic Information Technologies, Middle East Technical University, 06531, Ankara, 
Turkey - gsarp@metu.edu.tr 
Commission PS-13, WgS IV/3 
KEY WORDS: Remote Sensing; Road Extraction, Spatial Resolution, GIS, Ankara 
ABSTRACT: 
Road extraction from satellite imagery has become a heated research subjects in recent years. It is especially used in the city planning, 
cartography and to update previously detected roads in Geographic Information Systems (GIS) environment.In this study automated 
road extraction technique is applied to four different satellite images (SPOT, IKONOS, QUICKBIRD, ASTER) with different 
resolutions that belong to Ankara city of Turkey. The aim of this study is to find out the effect of the resolution on the automatically 
extracted roads. In this manner the roads are extracted from these four satellite images individually. Finally, the accuracy of 
generated results was tested with the GIS data layer that represents the reality. 
1. INTRODUCTION 
Many ways of road extraction have been proposed and they are 
quite different due to the differences instrategies, type and 
resolution of input images, experiment configurations, ways of 
processing and general assumptions, etc. (Fortier et. al., 1999). 
In the previous studies roads are extracted from satellite images 
using automated and semi-automated methods (Gruen, et.al., 
1997); (Baltsavias, et.al, 2001); (Wiedemann, C., Hinz, S., 
1999). The main advantages of automated road extraction are its 
ability to uniform approach to different images; processing 
operations are performed in a short time and its ability to extract 
roads which are not recognized by the human eyes. In semi- 
automated extraction method, roads are extracted from satellite 
image by using visual interpretation. There are several image 
enhancement techniques that can contribute to semi automated 
road extraction as filtering, image classification, band rationing 
etc/. Zheng et al. (1998) detected roads from satellite images by 
using filtering and edge detection processes. Ruisheng Wang 
and Yun Zhang, (2003) extracted roads from Quickbird images 
using classification techniques. 
Resolution of satellite images has important effect on roads or 
other objects to be discriminated. Images having different 
resolution include different types of road. In a low resolution 
images roads are exist as single line on the other hand in a high 
resolution images roads have particular width and the pixels 
located at two sides of road have irregular pattern because of 
existing of trees, cars and houses along the roads. 
Aim of this study is to detect the roads in urban areas from 
satellite images with different resolutions by using automated 
and semi-automated methods and to investigate the influence of 
resolution on road extraction. The automated road extraction in 
this study was performed by th^ LINE module of Geomatica 
software. The main idea is automatically detect a line of pixels 
as a vector element by examining local variance of the gray 
level in a digital image Koike et al. (1995). For the semi- 
automated extraction filtering and image classification 
procedures were applied. 
2. DATA AND METODOLOGY 
Study earned on western of Ankara, capital of Turkey, 
including 6 km 2 area (Figure 1). Five different satellite images 
with different resolutions (QUICKBIRD 2.4 m, IKONOS 4 m, 
SPOT-PAN 10 m, ASTER 15 m and LANDSAT-ETM 30 m.) 
were used and roads were tried to detect by using automated 
and semi-automated methods from these five images having 
different characteristics. 
Figure 1: Study area 
2.1. Automated Methods 
There are several different automated methods are used for 
extracting roads from images. Generally, automated road 
extraction consists of four steps; roads sharpening, roads finding, 
roads drawing and relating to extracted roads to each other. In 
this study automated road extraction carried out by using line 
module of PCI Geomatica software (Figure 2). 
According to automated extraction results, the roads through 
bare rural areas could be detected more easily and accurately. 
This is because of basic and regular structure of roads at bare 
rural areas, and they are far away from settlements. In a high 
resolution image detecting roads by automated module is hard. 
Because these roads have nonlinear and nonuniform structure,