AUTOMATIC REGISTRATION OF AIRBORNE IMAGE SEQUENCES 
BASED ON LINE MATCHING APPROACH 
ZHANG Pengqiang 3 ’ *, YU Xuchu 3 , HAN Li b , SONG Lihua 3 
a Zhengzhou Institute of Surveying and Mapping, Zhengzhou 450052, China - 
zpql978@163.com 
b Zhengzhou Institute of Surveying and Mapping, Zhengzhou 450004, China - 
WgS - PS: ICWG IIW 
KEY WORDS: Airborne remote sensing, Registration, Computer Vision, Feature extraction, Dynamic Change, Image sequences, 
Feature matching 
ABSTRACT: 
Traditionally, image registration approaches are classified to two categories: area-based methods and feature-based methods, where 
the latter use corresponding features as control objects. As the most popular feature in image processing, the line feature is well 
studied in recent years. In this article an automatic image sequences registration method is proposed which using corresponding 
straight line segments as control objects. According to the collinearity principle of corresponding straight lines, the image 
registration model is established. With this model, the straight line segment can be specified by arbitrary two points in it, but not its’ 
endpoints exactly. This method break out the limitation that corresponding features must be the same. Meanwhile, to realize 
automatic registration process of image sequences, the algorithms of automatic straight line segment extraction and automatic 
straight line segments matching are designed, which satisfy the demands of image registration, under the condition that the images 
have preliminary registered. 
1. INTRODUCTION 
Image registration approaches are classified to two categories: 
area-based methods and feature-based methods [Barbara 2003]. 
To estimate the geometrical transform parameters between the 
source image (slave image) and reference image (master image), 
area-based methods scan the whole parameter space and 
calculate the similarity between the reference image and the 
transform image which created from the source image, 
according to every possible set of parameters. Area-based 
methods are effective, if there exists only simple distortion 
between the source image and reference image, such as 
translation, similarity or affine distortion. The advantage is that 
the automatic registration process can be realized easily. 
However, if there exist complicated geometrical distortions 
between two images, the calculation increase rapidly as the 
increase of parameter dimensions and expanding of parameter 
values, so as to it’s too time-consuming to finish calculation in 
some conditions. 
midpoints of straight line segments and centroids of areas. In 
this way employed corresponding object features must be the 
same strictly. 
In recent years, some literatures from photogrammetry 
community have studied the feasibility to employ the advanced 
ground objects, for example, line objects, to automatic absolute 
orientation process. [Zhang Zuxun, 2004], [Zhang Zuxun, 2005] 
proposed generalized point photogrammetry, which introducing 
point, straight line, circle, arc, curve and limitless point, named 
as generalized points, into photogrammetric process. [Zhang 
Hongwei, 2004] demonstrated a concrete method to realize the 
automatic registration process between remote sensing image 
and vector map in detail, which using generalized points as 
control objects. In his method the geo-coded control generalized 
points are selected from vector map, while the corresponding 
features in image are extracted automatically. The generalized 
points are used as control objects to replace the ground control 
points (GCPs) in absolute orientation process. 
Ig|| 
In contrast, feature-based methods have more extensive 
adaptability and better stability, although their processes are 
more complicated. In these methods the object features which 
extracted from two images are employed as control objects. To 
estimate image registration model between the source image 
and reference image, feature-based methods calculate the 
geometry relationship of corresponding object features. 
According to the kind of employed object features, feature- 
based methods are classified to three classes: area feature-based 
methods, line feature-based methods and point feature-based 
methods. In practice, because of the difficulties in describing 
and matching of area features and line features, they are 
replaced by point features usually, such as endpoints and 
In this article, we introduce straight line into automatic 
registration process of airborne image sequences. According to 
the collinearity principle of corresponding straight lines, we 
establish the image registration model. With this model, the 
straight line can be specified by arbitrary two points in it, but 
not its’ endpoints exactly. This method break oui the limitation 
that corresponding lines and areas must be the same. 
Meanwhile, to realize the automatic registration process of 
image sequences, we design the algorithms of automatic 
straight line segment extraction and automatic straight line 
segments matching, which satisfy the demands of image