PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2017
109
Integrating Multiple Textural Features for
Remote Sensing Image Change Detection
Qingyu Li, Xin Huang, Dawei Wen, and Hui Liu
Abstract
This paper proposes a multi-texture change detection method
by integrating macro- and micro-texture features. Macro-tex-
tures are related to the information defined by the whole
image scene, while micro-textures describe distributions
and relationships of the gray levels within a local window.
Moreover, we propose two strategies, random forests (RF)
and a fuzzy set model, to integrate different characteristics
of the textures. Experiments were conducted on
ZY-3
(the first
civilian high-resolution stereo mapping satellite of China)
orthographic images of the cities of Wuhan and Tokyo, as
well as WorldView-2 multi-spectral images of the city of Kuala
Lumpur. Results showed that the wavelet-based features ob-
tained the highest accuracy among the macro-textures, while
the morphological attributes obtained the best results for the
micro-textures. By integrating both micro- and macro-tex-
tures, the texture combination using both RF and a fuzzy set
model can further improve the accuracy of change detection.
Introduction
Multi-temporal change detection is one of the most important
techniques for remote sensing applications. As a result, it has
been extensively applied in various aspects for dynamical-
ly monitoring and acquiring the trend of land evolution. A
large number of the existing approaches for remote sensing
change detection are based on spectral comparison between
multi-temporal images. These approaches, such as image
differencing (Al-Khudhairy, 2005), image rationing (Im and
Jensen, 2011), change vector analysis (
CVA
) (Chen
et al.
, 2003),
vegetation index differencing (Sohl, 1999), post-classification
comparison (Homer
et al.
, 2015), artificial neural networks
(
ANN
s) (Liu and Lathrop Jr., 2002), and support vector ma-
chine (
SVM
) (Zhang
et al.
, 2012), regard a single pixel as the
basic processing unit.
However, although the pixel-based and spectral-based
methods are widely used for remote sensing image change
detection, they have the following deficiencies:
1.
Pixel-based change detection neglects the neighboring
pixels and does not consider the spatial correlation
when determining whether or not a pixel has changed
(Johansen
et al.
, 2010).
Spectral-based change detection overlooks the textural and
structural characteristics of the images. When consid-
ering the spectral information alone, the results may
not be satisfactory for the change detection of remote
sensing data (Huang and Zhang, 2012), since the high
intra-class variation and low inter-class variation can
lead to inadequacy and uncertainty of the pattern anal-
ysis in the spectral domain (Huang and Zhang, 2013).
Furthermore, this problem becomes more serious when
processing high-resolution remote sensing imagery
(Huang and Zhang, 2013).
In this context, the textural information has recently been considered
in change detection in order to exploit spatial information to comple-
ment and enhance the spectral-based and pixel-based methods. Trans-
form-based textures are among the most commonly used textures for
change detection. In Klonus
et al.
(2012) and Aghababaee
et al.
(2012),
multi-temporal images were transformed by the use of a fast Fourier
transform, and the most suitable band-pass filter was applied to extract
the changed structures. The extended transforms, such as curvelet and
contourlet, have also been proved to have better shift-invariance property
and directional selectivity (Li
et al.
, 2014). Gray level co-occurrence
matrix (
GLCM
) has also been considered in texture-based change detec-
tion (He
et al.
, 2011). Moreover, in recent years, morphological textures
have attracted much attention for change detection, and it has been shown
that morphological operators are appropriate for describing the structur-
al change of high-resolution images by taking into account spatial and
structural correlation (Dalla Mura
et al.
, 2008). Another example of tex-
ture-based change detection is the fractal method. For instance, in Huang
et al.
(2011) and Aghababaee
et al.
(2012), the fractal method was
used for multi-temporal
SAR
change detection. Another fractal measure,
lacunarity, was also used to detect slum area change in Hyderabad, India,
using QuickBird and WorldView-2 imagery (Kit and Lüdeke, 2013).
Although a number of the existing papers have attempted
texture-based change detection, they only took a specific kind
of texture into consideration and neglected the comparison
and combination of different textures in change detection.
However, in fact, different textural features have different ef-
fects and performances in various image scenes. For instance,
transform-based methods describe the macro-textures, while
the
GLCM
and mathematical morphology focus on local struc-
tures. In this regard, one of the motivations of this study is to
better understand the texture-based change detection methods
by comparing and analyzing the performance and characteris-
tics of different textures for image change detection. More-
over, it can be stated that a mono-texture is biased for image
representation, as it is difficult to sufficiently describe the
complex and varied geospatial targets using only one texture
feature. Therefore, we propose to combine micro-textures (lo-
cal structures) and macro-textures (global characteristics) for
change detection, which has not been adequately addressed
in the current literature. In addition, it should be noted that,
in this research, a number of textures are exploited in change
detection and analysis for the first time, such as the three-di-
mensional wavelet transform (
3D-WT
).
Xin Huang is with the School of Remote Sensing and Informa-
tion Engineering, Wuhan University, Wuhan, China (corre-
sponding author: Xin Huang,
.
Qingyu Li, Dawei Wen, and Hui Liu are with the State Key
Laboratory of Information Engineering in Surveying, Mapping
and Remote Sensing, Wuhan University, Wuhan, China (cor-
responding author: Dawei Wen,
.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 2, February 2017, pp. 109–121.
0099-1112/17/109–121
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.2.109
