A Local Distinctive Features Matching Method for
Remote Sensing Images with Repetitive Patterns
Min Chen, Rongjun Qin, Haiqing He, Qing Zhu, and Xing Wang
Abstract
A novel feature matching method for remote sensing images
with repetitive patterns is proposed in this paper. Firstly, a de-
tector, with the feature response function considering geomet-
ric distinctiveness of image pixel as well as the support region
surrounding the pixel, is proposed to detect local distinctive
features. Secondly, those features with higher distinctiveness
are selected as seed points and matched. A matching reli-
ability indicator is proposed to select reliable seed matches.
Then, a coarse geometric transformation is computed based
on the seed matches to define a corresponding search area
for each feature. Finally, a feeble interest point searching
strategy is adopted to find correspondence for all the features.
Experimental results demonstrate that the proposed method
is able to obtain much more correct matches than traditional
methods, as well as the highest matching precision (around
90 percent) in the comparative evaluations for remote sensing
images with highly repetitive patterns.
Introduction
Remote sensing images from different platforms and their
processed products through photogrammetry and computer
vision techniques (e.g., geo-referenced/geo-registered images,
thematic maps, etc.) are nowadays widely used in many civil
and environmental applications, such as urban management,
forest monitoring, and precision agriculture. Image matching
as a fundamental step in the processing chain, is decisive to
the quality and success of the remote sensing products and
associated applications. Many image matching methods have
been proposed in the photogrammetry and computer vision
community, where the main efforts were posed on dealing with
geometric distortion and radiometric changes between images.
Despite the progresses in dealing with geometric and radio-
metric distortion between images, the focus of these methods
have not yet translated well to deal with difficult situations,
such as images with repetitive patterns. However, feature
matching on images with repetitive patterns is becoming an
unavoidable problem with the development of remote sensing
applications such as precision agriculture, ecological model-
ing, and desert mapping. These applications normally involve
large amount of images with highly repetitive texture patterns,
thus it is crucial to develop more robust feature matching
methods for reliable correspondence search in such images.
In this paper, a new matching method is proposed to
address matching points on images with highly repetitive
textures. Our approach is based on the key observation that if
the feature detector is guided by pixel matching potential (re-
flected by feature descriptor distinctiveness) across different
images, the detected features will be more prone to succeed
in matching. We consider feature detector and descriptor as
an integrated solution for point correspondence problem, in
which we devise a new local distinctive feature (
LDF
) detector,
a match reliability indicator, as well as a feeble interest point
set (
FIPS
) based matching strategy in our proposed workflow:
The
LDF
detector is designed to detect interest points on im-
ages with repetitive texture patterns by evaluating the distinc-
tiveness of both the pixel and the support region (surrounding
patches to compute feature descriptors) that contribute to
a point feature selection strategy that detects more unique
points. The matching reliability indicator is used to select
robust seed matches for building the geometric transforma-
tion for refined matching. The
FIPS
-based matching strategy
is adopted to reduce the probability of missing match while
reducing the search time of exhaustive method. Our main
contributions can be seen as three-fold.
First, we propose a new interest point detector considering
two levels of distinctiveness (i.e., pixel and support region) to
detect local distinctive features (
LDF
s). As compared to exist-
ing detectors that consider local maxima of gradients (either
or not in the scale space), the two levels of distinctiveness of
LDF
s offer a further differentiation in the local structure of a
region centered on the potential interest point, thus providing
much more information in evaluating the matching potential.
Second, we propose a matching reliability indicator that
incorporates both the distinctiveness of a pair of points and the
similarity of their feature descriptors to determine a match. Tra-
ditional descriptor similarity-based evaluation considers solely
the Euclidean distance between the feature descriptors. Given
that the feature distinctiveness are extracted within a local
region considering the minor differentiation of local textures,
our match reliability incorporating this feature distinctiveness
can eliminate false matches by selecting pairs both with similar
feature descriptors and high feature distinctiveness.
Third, a
FIPS
based matching strategy is adopted to reduce
the probability of missing match while guaranteeing time effi-
ciency. The points in the
FIPS
are normally defined as a sparse
point set and our matching strategy only scans over these
sparse points instead of exhaustive search, thus to provide
linear computation complexity.
The remainder of this paper is organized as follows: the next
Section introduces relevant works in the existing literature.
Min Chen and Qing Zhu are with the Faculty of Geosciences
and Environmental Engineering, Southwest Jiaotong University,
Chengdu, P.R. China (
).
Rongjun Qin is with the Department of Civil, Environmental
and Geodetic Engineering, The Ohio State University,
Columbus, OH; and the Department of Electrical and Computer
Engineering, The Ohio State University, Columbus, OH.
Haiqing He is with the School of Geomatics, East China
University of Technology, Nanchang, P.R. China.
Xing Wang is with the Key Laboratory for National Geographic
Census and Monitoring, National Administration of
Surveying, Mapping and Geoinformation, Wuhan, P.R. China.
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 8, August 2018, pp. 513–523.
0099-1112/18/513–523
© 2018 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.8.513
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2018
513
