PE&RS September 2014 - page 839

September 2014
Generation of Pixel-Level SAR Image Time Series
Using a Locally Adaptive Matching Technique
Liang Cheng, Yafei Wang, Manchun Li, Lishan Zhong, and Jiechen Wang
Synthetic Aperture Radar (
) image time series play an
important role in many applications. To construct pixel-level
image time series, we propose a locally adaptive image
matching technique for the high-precision geometric regis-
tration of
images. The basic idea is to adapt the local
characteristics of ground objects during the process of image
registration. Then, by analyzing the spatial distribution of
the error of each matched pair in the previous iteration, local
areas are divided based on the local clustering of pairs with
large errors. A new polynomial is then used to satisfy the
local geometric constraint. Based on this proposed matching
technique, we introduce a pixel-level
image time series
modeling method. The experimental results show that the av-
erage geometric error of corresponding pixels in this algorithm
is 0.073 pixels, while that of the
software is 0.242 pixels.
The Pearson correlation coefficients of 20 pixels’ time series
are above 0.85, indicating that the series bears high curve sim-
ilarity and geometric precision, which suggests the proposed
technique can provide high-quality
image time series.
Remote sensing time series information is a key component
of spatio-temporal data mining, and has particular theoretical
and practical value in the mining of massive and/or complicat-
ed multi-source remote sensing data. As a result, the modeling
of remote sensing time series has become an important topic of
research. At present, most of the widely-used remote sensing
image time series data are derived from
, Landsat, etc., which play important roles in
vegetation growth monitoring (Vancutsem
et al
., 2009; le Maire
et al
., 2011a; Fusilli
et al
., 2013), land cover classification (Ver-
et al
., 2008; Clark
et al
., 2010; Broich
et al
., 2011; Klein
et al
., 2012), and surface change monitoring (Verbesselt
et al
2010; Salmon
et al
., 2011; Li
et al
., 2012). In the field of micro-
wave remote sensing,
imaging can monitor the Earth’s sur-
face, irrespective of sunlight and weather conditions, and has
a limited capacity to penetrate certain surface objects. It also
benefits from multi-polarization, multi-angle incidence, and
multi-modality, affording it a unique advantage in monitoring
global change, as well as regional resources and environments.
Moreover, the ability of
imaging to overcome single-image
defects make it well suited to marine observation, disaster
monitoring, and a range of military applications.
In general,
image time series comprise three compo-
nents: (a) image-level time series, (b) region-level time series,
and (c) pixel-level time series. Compared with the image-level
and region-level time series, the pixel-level time series retain
more original information and detail of ground objects, allow-
ing for effective extraction of surface features and discovery of
rules for changes in ground objects. Unfortunately, due to
slant-range imaging, the radiometric intensity value cannot ac-
curately represent the backscattering coefficients of the ground.
Given the large changes in phase and radiation characteristics
of an
target over time, the usual speckling of
and the regular changes in
imaging mode and incidence
angle, the accurate acquisition of feature points can be very
difficult, resulting in poor image matching. Thus, high-preci-
sion radiometric correction and geometric matching play an
essential role in the construction of pixel-level
image time
Unlike optical images,
images are acquired by active
sensors, and have different imaging principles. They have vari-
able polarization (
) and different passes (descending
and ascending), which can change the intensity value of the
same pixel in two different images. Further, a degree of speck-
le noise is also inherent in
images. Optical images and
images have a different appearance, and reveal different
characteristics of the imaged area. For these reasons, existing
registration algorithms for optical remote sensing images can-
not be directly used in the registration of
To achieve pixel-level
image time series, we propose a
locally adaptive image matching technique for the high-pre-
cision geometric registration of
images. A polynomial is
used as the geometric constraint model of master-slave image
matching, and its parameters are computed to register the mas-
ter-slave pairs. By analyzing the spatial distribution of the error
of each pair, local areas containing large matching errors are
divided based on the local clustering of pairs, and a new poly-
nomial is then used to satisfy the local geometric constraint.
This is the basic idea of the proposed locally adaptive matching
technique. Binary partitioning is employed in the iterative pro-
cess to gradually adapt to the local characteristics of the ground
objects and achieve high-precision geometric registration. In
addition, based on this proposed matching algorithm, we intro-
duce a pixel-level
image time series modeling method.
Related Work
Construction of
Image Time Series
Remote-sensing time series comprise three types: image-level,
region-level, and pixel-level. Image-level remote sensing
Jiangsu Provincial Key Laboratory of Geographic Information
Science and Technology, Nanjing University, 163 Xianlin Av-
enue, Nanjing 210023, China; Collaborative Innovation Center
for the South Sea Studies, Nanjing University, Nanjing Univer-
sity, 163 Xianlin Avenue, Nanjing 210023, China; Department
of Geographic Information Science, Nanjing University, 163
Xianlin Avenue, Nanjing 210023, China (
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 9, September 2014, pp. 839–848.
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.9.839
811...,829,830,831,832,833,834,835,836,837,838 840,841,842,843,844,845,846,847,848,849,...914
Powered by FlippingBook