ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision“, Graz, 2002 
  
SEMI-AUTOMATIC ROAD EXTRACTION FROM 
HIGH-RESOLUTION SATELLITE IMAGE 
Huijing Zhao, Jun Kumagai, Masafumi Nakagawa, Ryosuke Shibasaki 
University of Tokyo 
Commission III 
KEY WORDS: Road Extraction, High-Resolution Satellite Image, Urban area, Semi-automatic 
ABSTRACT: 
In this research, a method is proposed to create and/or update road maps in urban/suburban area using high-resolution satellite 
images. “Road mask” is defined in this research as a mask of road pixels, which are discriminated from others using a commercial 
remote sensing software. "Road seed" is defined in this research as a directional point, indicating that a road is passing through the 
point along the direction. Road seeds are extracted from edge pixels. Road line extraction is conducted in a semi-automatic way by 
fusing both road mask and road seeds. Experiments are conducted using an IKONOS image of nearby KAWAGOE city, Japan, with 
a ground resolution of 1 meter, and four bands, i.e. red, green, blue, and near infrared. Experimental results show that the method is 
valid in extracting main roads in high dense building area and all roads in countryside efficiently. 
1. INTRODUCTION 
Since the launch of commercial satellites, such as IKONOS and 
QuickBird, high-resolution satellite imageries at the resolution 
close to that of aerial photograph are available periodically. One 
of the major expectations is in the use of updating 
urban/suburban maps, such as road network for car navigation 
system and other GIS applications. In this research, we propose 
a semi-automatic method for generating new and/or updating 
existing road maps of urban/suburban area. 
1.1 Previous Works 
Up to now, numerous methods have been proposed for the 
extraction of road features from space imagery. 
Barzohar and Coopper 1996 proposed an automatic method of 
extracting main roads in aerial images. The aerial image is 
partitioned into windows, road extraction starts from the 
window of high confidence estimates, while road tracing is to 
perform a dynamic programming to find an optimal global 
estimate. Geman and Jedynak 1996 proposed a semi-automatic 
method, where given a start point and a start direction, a road is 
extracted from a panchromatic SPOT satellite image by playing 
"tests" about the "true hypothesis". Gruen and Li 1997 
formulated the problem using an active contour model in a least 
square context (LSB-Snake), where given a number of seed 
points, an initial road template is first generated then adjusted to 
optimised an energy function on both photometric and 
geometric characteristics. Fiset ef al. 1997 proposed a map- 
guided method to update the map of road network using SPOT 
imagery. Latest research efforts can be found in Park and Kim 
2001, where a semi-automatic road extraction is proposed using 
template matching. 
Most of the existing methods are based on a road model, where 
the roads are assumed to follows a number of generalities. For 
example, in Barzohar and Coopper 1996, assumptions on a 
geometric-stochastic road model are clearly listed as follows. 
1) Road width variance is small and road width change is 
likely to be slow. 
2) Road direction changes are likely to be slow. 
3) Road local average grey level is likely to vary only slowly. 
4) Grey level variation between road and background is likely 
to be large. 
5) Roads are unlikely to be short. 
However they are not always true, as road images vary a lot 
with ground resolution, road type, density of surrounding 
objects and so on. A specific road model as well as a road 
extraction method is required for extracting road lines using 
high-resolution satellite image, e.g. IKONOS and QuickBird 
images, where a road model is preferable to have as less but 
generic assumptions as possible. 
1.2 Outline of the Research 
In this research, we propose a semi-automatic method of road 
extraction from urban/suburban scene using high-resolution 
satellite images, which have a ground resolution of about 1 
meter, and four bands, i.e. red, green, blue, and near infrared. 
The method consists of three steps as shown in Figure 1. 
A “road mask” is defined in this research as a mask of road 
pixels, which are generated by classifying road pixels of a 
multi-spectrum satellite image using a commercial remote 
sensing software. 
A “Road Seed” is defined in this research as a directional point, 
indicating that a road is passing through the point along the 
direction. Road seeds are extracted by tracing edge pixels, as a 
long edge line with only a slow change of direction suggest a 
road or river passing through. 
A - 406