SEMI-AUTOMATIC EXTRACTION OF RIBBON ROADS FORM HIGH RESOLUTION 
REMOTELY SENSED IMAGERY BY COOPERATION BETWEEN ANGULAR 
TEXTURE SIGNATURE AND TEMPLATE MATCHING 
X. G. Lin a ’ b > *, J. X. Zhang 3 , Z. J. Liu 3 , J. Shen b 
a Chinese Academy of Surveying and Mapping, Beijing 100039, China - linxiangguo@gmail.com, - 
(zhangjx, zjliu)@casm.cn 
b School of Resources and Environment, Wuhan University, Wuhan 430079, China 
Commission III, WG III/5 
KEY WORDS: Road extraction; Semi-automatic; Angular texture signature; Template matching 
ABSTRACT: 
Road tracking is a promising technique to increase the efficiency of road mapping. In this paper an improved road tracker, based on 
cooperation between angular texture signature and template matching, is presented. Our tracker uses parabola to model the road 
trajectory and to predict the position of next road centreline point. It employs angular texture signature to get the exact moving 
direction of current road centreline point, and moves forward one predefined step along the direction to reach a new position, and 
then uses curvature change to verify the new added road point whether right enough. We also build compactness of angular texture 
signature polygon to check whether the angular texture signature is suitable to be used to go on tracking. When angular texture 
signature fails, least squares template matching is then employed instead. Cooperation between angular texture signature and 
template matching can reliably extract continuous and homogenous ribbon roads on high resolution remotely sensed imagery. 
1. INTRODUCTION 
Extraction of road from digital aerial/satellite imagery is not 
only scenically challenging but also of major importance for 
spatial data acquisition and update of geodatabases. Traditional 
manual plotting is time consuming and expensive, so automatic 
acquisition and update of road data is greatly needed. In (Bajcsy 
and Tavakoli, 1976; Wang and Newkirkr, 1988; Trinder and 
Wang, 1997; Long and Zhao, 2005; Haverkamp, 2002; Hinz 
and Baumgarter, 2003; Zhang and Couluigner, 2006; Barzohar 
and Cooper, 1996; Gardner and Roberts, 2001; Baatz and 
Schape, 2004), various fully automatic approaches are proposed. 
But the road characteristics vary considerably with ground 
resolution, road type, density of surrounding objects, and light 
conditions and so on, adding that the limits of state of the art on 
computer vision and photogrametry, the desired fully 
automation could not be achieved by now, however, semi 
automatic approach that retains the human operator in the loop 
where computer are used to assist human performing is 
considered to be a good compromise between the fast 
computing speed of a computer and the efficient interpretation 
skills of an operator. And quite a lot of promising approaches 
for semi-automatic road extraction have been proposed in the 
last two decades. Quam (1978) tracked road by road surface 
model and profile model; Nevada and Babu (1980) proposed 
edge-based technique; Mckeown and Denlinger (1988) 
combined edge-based and profile correlation based approach; 
Vosselman and de Knecht (1995), Baumgartner (2002) and 
Zhou (2006) used least square profile matching; Park and Kim 
(2001), Hu, Zhang and Zhang (2000) employed template 
matching; Grun and Li (1995), Merlet and Zerubia (1996) 
connected road seeds by dynamic programming; Grun and Li 
(1997) used snakes to optimize the path of road seed points; 
Vandana and ChandaraKanth and Ramachandran (2002) 
employs minimum cost to follow a path; Baltsavias (2004) 
revised road map based on existing geodata and knowledge. But 
a standard cliché of road extraction is that every algorithm has 
its limits, so we believe that a number of techniques developed 
for different classes of road will lead to a many-branched 
solution for road extraction that will be effective for a wide 
range of road types. Improved angular texture signature is 
proposed and cooperation between angular texture signature 
and template matching is employed to semi-automatically 
extract road network in this paper. 
Road characteristics and the principles of the proposed 
algorithm are described in Sect. 2. In Sect. 3 we introduce the 
process of our tracker. Section 4 compares our tracker with 
classic algorithms. Section 5 evaluates the tracker by a case 
study. Section 6 summarizes the results of our study and makes 
a conclusion.