Road Extraction in Urban and Suburban Area Using a Contextual Method 
Tian Lianghu 
Lecturer, Department of Earth Science 
Zhejiang University, Hangzhou, China. 
Commission III 
Abstract ; 
Linearment enhancement is one important subject of re- 
mote sensing. Many methods and algorithms for linear— 
ment enhancement have already been presented and de- 
signed which generally use the abrupt change of bright- 
ness on the edge of linearments. These methods, howe- 
ver, have one shortcoming that they are not efficient if 
much more noises exist in the image or there are less 
brightness change between the background and linear- 
ment. This paper presents a contexual method to extract 
urban road and suburban railway in Yueyang district of 
Hunan Province using the shape information. It is 
known that roads are such linearments that they are 
overlayed on a homogeneous background which differs 
from other linearments like the edge of two bodies. This 
enables the clustering technique instead of general en- 
hancing methods to conveniently and efficiently enhance 
Keywords. Pattern Recognition, Feature Extraction, 
1. INTRODUCTION 
The linear feature is the one important characteristic on 
the remote sensing image. Linearments on the image 
contain mainly two types. One is the edge boundary be- 
tween two regions of different constant grey level. The 
ideal step edge has the cross section shown in fig. 1. An- 
other is the linear object which occurs in a large homoge- 
neous object such as road. Fig. 2 and fig. 3 show the ba- 
sic linear objects on the images. In real conditions, these 
linear objects always present the case as fig. 4. 
434 
the road information in which much more other non — 
road noises are contained. Then contexual information , 
the ratio of the length to width, is used to eliminate the 
non —linearment noises. À mass centering method is de- 
veloped to connect the continual points on the road lin- 
earment which can form a complete line and has no ef- 
fect on the adjacent points. Another algorithm is de- 
signed to extract the length and width of the line in 
which the continuity is considered that differs from gen- 
eral convolution technique. The ratio of the extracted 
length and width is used as contextual information to re- 
move the other non— road pixels. The result shows that 
this method is considerablely efficient to extrat the road 
features in the urban and suburban area on the remotely 
sensed data. Furthermore, it is helpful for monotoring 
the urban road distribution and the urban expansion. 
Classification, Algorithm 
The detecting and extracting of the linear features on the 
remote sensing image have been studied by many scien- 
tists and experts. Various methods and approaches have 
been developed which use diverse masks covered on the 
image to detect abrupt changes of grey level. These ap- 
proches can be used to detect the roads in the case dis- 
cussed above. But in many cases, it exists great varieties 
of noises and the roads can not effectively be extracted. 
It is obvious that previous methods developed for edge 
detection are used only on single band image data. They 
have 
impo 
road 
grou 
In tk 
sual 
tures 
and | 
also 
foun 
denti 
the | 
prov 
pute: 
ed t« 
The 
Prov 
ing I 
its n 
is th 
tion. 
Yue