1157 
AN INTEGRATED FEATURE BASED METHOD FOR SUB-PIXEL IMAGE MATCHING 
Weili JIAO, Yaling FANG, Guojin HE 
Center for Earth Observation and Digital Earth, Chinese Academy of Sciences 
P. O. Box 2434, Beijing 100086, CHINA 
wljiao@ceode.ac.cn 
KEY WORDS: Feature-Based Method, Sub-pixel Accuracy, Surface Fitting, Image Matching 
ABSTRACT: 
Feature matching is to find the pair-wise corresponding features in the reference and sensed images. The features can be an edge, a 
comer, an end point, a line or a curve, etc. Image matching as the fore-step of image fusion and registration, image mosaic, 
automatic change detection, etc., the matching accuracy highly impacts these applications. Normally these applications require the 
matching accuracy to be sub-pixel. However, the matching unit of the feature-based methods is “one pixel”. In this paper, after 
analyzing the image matching algorithm, an integrated method is proposed to detect the point features by the means of the combined 
advantages of Harris operator and Forstner operator. With this method, the matching accuracy can reach sub-pixel. The thought 
of matching principle based on bridge mode method on initiate matching points is introduced, when matching feature points by the 
method of template matching and conformal transform, to eliminate the low precision effect on the following matching points. The 
effect is caused by the cumulated error due to the low precision of initiate matching points. The proposed method has been tested 
by three pairs of overlap SPOT images. These images were captured in the flat area, mountainous area, and in different time. The 
results show that the matching accuracy can reach sub-pixel, and the position is accurate with the method described in this paper. 
The method is also robust, effective, and suitable for realization of automatic image matching. 
1. INTRODUCTION 
Image matching is finding correspondences between tow 
elements of images. It is widely used in remote sensing, 
medical imaging, computer vision etc. The applications in 
remote sensing include image fusion and registration, image 
mosaic, automatic change detection, environment monitoring, 
weather forecast, super resolution image creation, etc. 
The common image matching methods can be divided into 
intensity-based methods and feature-based methods (Ni, and 
Liu, 2004). The widely used methods are feature-based methods. 
Feature-based matching methods are typically applied when the 
local structural information is more significant than the 
information carried by the image intensities. They allow 
registering images of completely different nature and can 
handle complex between image distortions (Zitova and Flusser, 
2003). 
The feature-based methods are to detect two sets of features in 
the reference and sensed images. The features can be an edge, 
a comer, an end point, a line or a curve, etc. Feature matching 
is to find the pair-wise corresponding features. Image 
matching as the fore-step of image fusion and registration, 
image mosaic, automatic change detection, etc., the matching 
accuracy will impact these applications. Normally these 
applications require the matching accuracy to sub-pixel. The 
matching unit of the feature-based methods is “one pixel”. 
The pair-wise corresponding features can be used as an input 
for sub-pixel matching with other methods. 
In this research, after analyzing several kinds of feature interest 
operators, an integrated feature based method is proposed for 
sub-pixel image matching. Firstly, Harris operator is used for 
detecting feature points. Each point is corresponding to one 
pixel. So the matching accuracy can only reach one pixel. 
Then the detected points by Harris operator is regarded as the 
window center of Forstner operator. The surface fitting 
algorithm is used to calculate the more accurate position of the 
feature. With this method, the matching accuracy can reach 
sub-pixel. During feature matching by the method of template 
matching and conformal transform, the thought of matching 
principle based on bridge mode method on initiate matching 
points was introduced, to eliminate the low precision effect on 
the following matching points. The effect is caused by the 
cumulated error due to the low precision of initiate matching 
points. 
2. FEATURE-BASED MATCHING METHODS 
2.1 Feature Extraction 
The feature extraction methods are to detect two sets of features 
in the reference and sensed images. The features can be points, 
lines or regions. In this research the feature detection methods 
are mainly deal with the point features. The point feature’s 
group consists of methods working with line intersections, road 
crossing, centers of regions, end points, comers, etc. Much 
effort has been spent in developing precise, robust, and fast 
method for comer detection. Point feature detection operator 
is also called interest operator. The reputable operators are 
Harris, Forstner, Moravec, etc. The integration of Harris and 
Forstner operators are used for the detection of point features in 
this paper. 
2.1.1 Harris operator (Harris and Stephens., 1988) 
It computes a matrix related to the autocorrelation function of 
the image. The squared first derivatives of the image signal 
are averaged over a window and the eigenvalues of the resulting 
matrix are the principal curvatures of the auto-correlation 
function. An interest point is detected if the found two 
curvatures are high. The main advantages of Harris operator