IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 7, NO. 3, JULY 2010 491
Remote Sensing Image Registration Based on
Retrofitted SURF Algorithm and Trajectories
Generated From Lissajous Figures
Zhi Li Song and Junping Zhang, Member, IEEE
Abstract—In this letter, we propose a novel remote sensing
image registration method by optimizing the Speeded Up Robust
Features (SURF) and developing a new similarity measure func-
tion based on trajectories generated from Lissajous figures. Com-
pared with SURF which has a low feature-matching rate in some
complex cases, the retrofitted SURF algorithm is more robust and
accurate. The algorithm greatly improves the correct matching
rate to over 80%. Furthermore, the recognition capability of the
similarity measure is enhanced by using a trajectory disturbance
strategy, which is a significant displacement in the trajectory
induced by a minor error of the transformation parameters. Ex-
periments show the promising performance of the proposed image
registration method.
Index Terms—Image registration, remote sensing, Speeded Up
Robust Features (SURF).
I. INTRODUCTION
R
EMOTE sensing images are mostly multimodal, which
means that a collection of images of the same scene are
sampled from different sensors and/or taken at different times.
To analyze such images, it is necessary to register the images
and extract some crucial features from them [1]. However, the
performance of the registration algorithms suffers from two
issues. First, there always exist geometric distortions stemmed
from rotation, scale, affine transformations, etc. Among them,
affine transformation is often used to approximate local trans-
formation when the depth of an object is much smaller com-
pared with the viewing distance. Second, due to considerable
diversities of sensors, multiple pixel intensities in an image may
correspond to a single pixel intensity in another. Consequently,
features in an image may partially appear or even disappear at
all in another one.
One feasible way to solve the first issue is to extract features
from the reference and sensed images separately, followed by
picking a descriptor to find out an appropriate relation between
the two sets of features [2]. If the descriptor is independent,
stable, and invariant to image transformation, then the image
Manuscript received May 14, 2009; revised September 2, 2009 and
November 1, 2009. Date of publication March 4, 2010; date of current version
April 29, 2010. This work was supported in part by the 973 Program under
Grant 2010CB327900 and in part by the National Natural Science Foundation
of China under Grant 60975044.
Z. L. Song is with the School of Computer Science, Fudan University,
Shanghai 200433, China (e-mail: zlsong@fudan.edu.cn).
J. Zhang is with the Shanghai Key Laboratory of Intelligent Information
Processing and the School of Computer Science, Fudan University, Shanghai
200433, China (e-mail: jpzhang@fudan.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LGRS.2009.2039917
registration will attain a reliable performance [3]. In the last
few decades, many feature-based methods devote to extracting
key features such as point, line, edge, region, and so on [4],
[5]. Among them, the scale-invariant feature transform (SIFT)
algorithm is to detect, describe, and match such local features
[6]. With the help of 2-D Haar wavelet responses, integral
images, and scale space technique, the Speeded Up Robust
Features (SURF) is faster and more robust against different
image transformations than SIFT [7]. However, they are only
invariant to minor affine transformation. Furthermore, they are
all sensitive to image multimodality.
To address the second issue, it is necessary to develop a
similarity measure to enhance the robustness and improve the
accuracy of image registration under multimodalities. Cur-
rently, most of the known similarity measures are area based,
such as intensity-based [1], [8], frequency-based [9]–[11], and
mutual-information (MI)-based [12], [13] techniques. Note that
the MI is regarded as an efficient similarity measure in the mul-
timodal image registration [12], [13]. Given two discrete ran-
dom variables X and Y with marginal probability distributions
p(x) and p(y) and joint probability distribution p(x, y), specif-
ically, the MI that measures statistical dependence between
the two variables X and Y is defined as I(X, Y )=H(X)+
H(Y ) − H(X, Y ), where H(Z)=−
z∈Z
p(z) log p(z) and
H(X, Y )=−
x∈X
y∈Y
p(x, y) log p(x, y).
Considering the aforementioned pros and cons, we thus pro-
pose a novel image registration approach based on Retrofitted
SURF algorithm and TRajectories generated from Lissajous
figures (RSTR). The main contributions of the proposed RSTR
algorithm include the following: 1) We obtain a higher cor-
rect matching rate by retrofitting the SURF algorithm; 2) we
propose an effective criterion to identify the correctness of
feature matching; 3) we achieve good performance of image
registration by defining a new similarity measure based on
trajectories generated from Lissajous figures; and 4) with the
help of a trajectory disturbance strategy, the accuracy of the
similarity measure is greatly improved.
II. P
ROPOSED RSTR ALGORITHM
In this section, we will introduce the proposed RSTR algo-
rithm in detail.
A. Extracting Parameters of Local Transformation From
TAR Signatures of Curves
First, in order to achieve a coarse transformation, we obtain
a collection of matched feature point pairs by SURF, ranking
1545-598X/$26.00 © 2010 IEEE