Ill 
STUDY ON A MATCHING METHOD OF AIRBORNE SAR IMAGES 
FAN Hong-dong1 *, DENG Ka-zhong 
College of Environment and Spatial informatics, CUMT, Xuzhou, Jiangsu 221008, China - cumtfhd@163.com 
Commission YD, WG V1I/2 
KEY WORDS: SAR; Matching; Algorithms; Edge; Processing; Method 
ABSTRACT: 
As an important step and technical nodus of synthetic aperture radar (SAR) image applications, image match is an important process 
in which the corresponding relations of pixel points between many SAR images can be got. Usually, the accuracy of image match 
directly affects the next operations of SAR images. Nowadays, image match methods can be mainly divided into two kinds: one is 
based on region and the other one on characteristic. The former usually uses normalization cross-correlation as a matching measure, 
so these methods aren’t good enough to meet the requirements of SAR images or multisensor image match. And the latter mainly 
uses common characteristics from two images to make image match. Generally speaking, point characteristics are affected by noises 
much more. Given the coherent speckle noises of SAR images, it is better to make image match using edge or outline characteristics. 
Considering these problems, a new match method and flow of SAR image pairs based on edge detection and “virtual epipolar” are 
proposed and researched in this paper. First of all, a new edge detector, combined by Canny operator and Ratio operator, is used to 
generate edge information. In order to improve the matching speed and quantitatively analyze the results, some edge points with 
more obvious characteristics are obtained by Moravec operator which can be regarded as a base points set. Next, in order to find the 
corresponding points faster, a search strategy based on “virtual epipolar” is proposed and established. In the third step, using edge 
information combined with gray-level information of the image as the matching measure, the corresponding points are searched 
along the “virtual epipolar” line. To get rid of the wrong corresponding points, a quadratic polynomial method is used. At last, 
experiments have been conducted on airborne SAR image pairs with lm resolution in Chengdu, China. As a result, the match 
method based on edge detection and “virtual epipolar” line studied in this paper is feasible, and the matching accuracy and velocity 
have improved much more than some other methods. 
1. INTRODUCTION 
Synthetic aperture radar (SAR) -commonly known, is an active 
microwave imaging system. Due to its working capability in all 
weather and all time, it is much more suitable for topographic 
mapping and targets detective in such areas with atrocious 
weather. In order to get more information from SAR images, a 
lot of SAR image processing methods have been developed. As 
an important step and technical nodus of SAR image 
applications, image match is a process in which the 
corresponding relations of pixel points between many SAR 
images can be got. The accuracy of image match directly 
affects the next operations of SAR images. Nowadays, image 
matching methods can be mainly divided into two kinds: one is 
based on the region and the other one on the characteristic 11 1 
Generally speaking, SAR image pairs are possibly captured in 
different time, with different angles of view and working bands, 
and even different carriers. Therefore, there are more speckle 
noises and bigger geometric distortions in SAR image pairs, for 
which the regional matching method is not suitable. However, 
the latter method that mainly uses common characteristics of 
images (for example: edge information, shape information, 
feature points and so on) to make image matching is much more 
suitable for SAR image pairs. Considering these problems, a 
new match method and flow of SAR image based on edge 
detection and “virtual epipolar” are proposed and researched in 
this paper. The main contents can be described as follows: (1) A 
new edge detector, combined by Canny operator and Ratio 
operator, is used to generate edge information; (2) In order to 
find the corresponding points faster, a search strategy based on 
“virtual epipolar” is proposed and established; (3) To get rid of 
the wrong corresponding points, a quadratic polynomial method 
is proposed. 
2. EDGE DETECTION 
Due to the particular imaging mechanism, there are more 
differences in SAR and optical image. Generally, in SAR image, 
the signal strength is larger which can induct more 
multiplicative noises. Even in the homogeneity areas (rice 
paddy, dry land, etc.), the reflecting signal intensity of same 
object in different position is usually different. Therefore, the 
classical edge detection operators, such as Roberts, Laplacian, 
Canny and so on, are not suitable for SAR image. Ratio 
operator, as the most typical edge detection algorithm of SAR 
image, which regard the regional gray average as the detecting 
character value, can availably avoid the disturbance of noises 
and get a higher precise, but the edge line is wider 121 . Therefore, 
in this paper, a combination of Ratio and Canny operator is 
made to improve the edge detector of SAR image. 
2.1 Canny Detector 
In 1986, Canny proposed three criterions to estimate the edge 
detector 131 : (1) low error rate, namely, should not leak edge 
pixel and also not obtain non-edge points; (2) high precise of 
detecting position, that is, the edge points detected should be in 
* Corresponding author.