535 
RESEARCH ON ROAD EXTRACTION SEMI-AUTOMATICALLY FROM 
HIGH RESOLUTION REMOTE SENSING IMAGES 
Haitao ZHANG, Zhou XIAO, Qing ZHOU 
Beijing Institute of Surveying and Mapping, 15, Yangfangdian, Haidian District, Beijing, China 
- zhanght@bism.cn 
KEY WORDS: Extraction, Road, Semi-automation, High resolution, Remote Sensing, LSB-Snake Model, Template Matching 
ABSTRACT: 
Nowadays, there have substantive research on road extraction automatically form RS images, restricted by low understanding level 
of images, the automatically extracting method is not robust, and a great many errors existed. The LSB-Snake model is an effective 
method to extract linear object semi-automatically, but needs manual input of road character for extraction, and not robust while the 
initial seed points are not dense enough. These hold down the working efficiency of LSB-Snake model. This paper put forward an 
auto-initial-valued LSB-Snake model, which use self-adapt template matching method to provide the road character to LSB-Snake 
model, and add seed points based on the initial points at the same time automatically. Experiments indicate: Given the same amount 
of initial seed points, our method is more robust than LSB-Snake model; Needn’t manual input the road character, the 
auto-initial-value LSB-Snake model is more automatic than LSB-Snake model; The auto-initial-value LSB-Snake model can 
overcome the shade or shelter of land objects such as building and trees, and more powerful in anti-jamming than LSB-Snake model. 
The methods this paper put forward can extract road feature from remote sensing image efficiently. 
1. INTRODUCTION 
Linear features extraction from Remote Sensing (RS) images is 
a frequently researched issue, and road feature is one of the 
most important features in RS images. Nowadays, there have 
substantive research on linear feature extraction form RS 
images, such as the multi-scale space strategy method 11] , the 
perspective group method 121 , the manual neural network 
method 121 , the classification method, the active contour model 
(Snake model) method 14,51 , the template matching method 16,7,81 , 
etc. These extraction methods can be classified by 
semi-automatic and automatic. Many researchers are interested 
in the automatic extraction method, although it has been 
researched for about thirty years, no system presented put into 
application robustly without human interfere, and the 
semi-automatic extraction method such as template matching 
and Snake model are more applicable. 
Road is a linear feature, and one of the most important feature 
in RS image, the extraction of road is important for urban layout, 
communication and surveying, etc. 
This paper put forward a self-adapt template matching method 
to match the road in RS images by templates of series width, 
which is named as self-adapt template matching method, then 
takes the result of the template matching into LSB-Snake model 
as initial value to extract road from RS images, which is named 
as Auto-initial-valued LSB-Snake model. The flowchart of our 
method as follows. 
High Resolution 
Remote Sensing Image 
Seed Points 
1 
Self-adapt Template 
Matching Method 
Road Information 
JL 
Dense Seed points 
Based on the template matching method, this paper put forward 
a method of self-adapt template matching, using the result of it 
as the initial value of LSB-Snake model, named 
“auto-initial-valued LSB-Snake model” to extract linear road 
feature from high resolution RS images semi-automatically. 
i 
Auto-initial-valued 
LSB-Snake Model 
2. AUTO-INITIAL-VALUED LSB-SNAKE MODEL 
Road is a kind of typical linear-like feature, it’s character can be 
concluded as gray scale, geometry, topology, function and 
conjunction or context obligation etc. among them, gray scale is 
the most important character, the gray scale of a road can be 
expressed as linear features of some the bright contrast between 
edge and the middle of road, the gray scale template of road’s 
section is designed according to the gray scale character of road. 
We match the road in RS image using road section template. 
I 
Extracted Road 
Fig. 1 Flowchart of Road Extraction Method 
Self-adapt template matching method and auto-initial-valued 
LSB-Snake model are the key technique of the paper.