Verbal Non Verbal
Verbal Non Verbal
Verbal Non Verbal
|
|
.
|
\
|
=
i
T
i
T i
SR
X
X
Y
, i=1, 2,.N (2)
It reflects the change information through
the ratioing of two images. After the absolute value
operation, the pixels with larger values correspond
to higher possibilities of change, and those pixels
with a value close to 0 are hints to no-change areas.
Ratioing is immune to false positives caused by sun
elevation angle, shadows and terrain, at least to
some extent.
1.1.3 Absolute distance
The ability to determine absolute distance
to an object is one of the most basic measurements
of remote sensing. It is denoted by Y
AD
:
=
=
N
i
i
T
i
T AD
X X Y
1
1 2
, i=1, 2,.N (3)
It integrates the change information on
each band into a single band through simple
addition.
1.1.4 Euclidean distance
It is also known as minimum distance. It is
denoted by Y
ED
:
( )
=
=
N
i
i
T
i
T ED
X X Y
1
2
1 2
, i=1, 2,.N (4)
It calculates the mean spectral reflectance
in each band for each class. Distance of each image
pixel to class means is calculated. Pixel is assigned
to class for which minimum distance is smallest. In
some cases a maximum distance may be provided
(not in Multi-spectral image) Pixels exceeding
maximum distance are placed in other category.
1.1.5 Chi Square transformation
Y is distributed as a Chi-square random
variable with degrees of freedom ( is the number
of bands)It is denoted by Y
CST
:
=
|
|
.
|
\
|
=
N
i
diff
i
i
T
i
T
CST
X X
Y
1
2
1 2
o
, i=1, 2,.N (5)
Where
i
diff
represents the standard
deviation. Multiple bands are simultaneously
considered to produce a single change image.
S.Anita Johannah Golda, B.Bala Murugan / International Journal of Engineering Research
and Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 2, March -April 2013, pp.1546-1551
1548 | P a g e
Fig.1. Flowchart of the fusion procedure aimed at fusing multiple spectral change difference (SCD) images
1.2 Feature level fusion:
Medium-level fusion, also called feature-
level fusion, uses attributes (or features) extracted
from the raw data as inputs and uses them into new
features, or feature map. Several features which are
extracted from the raw data are used for tracking.
Feature-level fusion approaches for target tracking
can be formulated using the powerful framework of
graphical models. The goal of dissertation is to
develop feature-level information fusion methods
for target tracking in High Speed Networks.
In the feature-level fusion, each sensor
observes an object, and a feature extraction is
performed to yield a feature vector from each
sensor. After using an association process to sort
feature vectors into meaningful groups, these
feature vectors are then fused and an identity
declaration is made based on the joint feature
vector. The fusion model which is used in feature
level fusion is
1.2.1 Fuzzy Set theory(FS)
The use of fuzzy sets here is consistent
with that of fuzzy classification methods that are
increasingly used in remote sensing to allow pixels
to retain some degree of membership in one or
more land use classes [8].A fuzzy set is one for
which the degree of membership for any element of
the set may range from zero to one, and so is well
suited to ambiguous or partial membership. Fuzzy
set theory defines set membership as a possibility
distribution. The general rule for this can expressed
as:
] 1 , 0 [ ] 1 , 0 [ :
n
f (6)
Where n some number of possibilities. This
basically states that we can take n possible events
and use f to generate as single possible outcome.
1.3 Decision level fusion:
High-level fusion, also called decision-
level fusion, takes the decisions from each sensor
as inputs and fuses them to obtain a global decision.
In the decision-level approach, each sensor
performs independent processing to produce a
declaration of identity, and then the declarations of
identity from each sensor are subsequently
combined via a fusion process. The fusion models
in decision level fusion are
1.3.1 Majority voting(MV)
It is a basic and simple decision
integration method [9], designed to combine the
output results by multiple processors. The main
idea of is to arrange the identified results according
to some specific voting rules, for example, simple
Compute SCD datasets
Thresholded image
Feature level fusion
Binary image
Decision level fusion
Change map Change map Change map Change map
Same?
Give images of same size
Image at
time T1
Image at
time T2
Compare size
Kittler-Illingworth
segmentation algorithm
Auto-thresholding
Fuzzy Set theory Majority voting Dempster Shafer
Fuzzy Integral
Yes
No
S.Anita Johannah Golda, B.Bala Murugan / International Journal of Engineering Research
and Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 2, March -April 2013, pp.1546-1551
1549 | P a g e
majority voting and weighted voting rules.
1.3.2 Dempster-Shafer(D-S) evidence theory
This method utilizes probability intervals
and uncertainty intervals to determine the
likelihood of hypotheses based on multiple
evidence. In addition, it computes likelihood that
any hypothesis is true. The concept of using a
Dempster-Shafer approach is to fuse multi-sensor
data. Dempster-Shafer evidence theory allows the
representation of both imprecision and uncertainty
and it is well suitable for remote sensing data
fusion problems.
1.3.3 Fuzzy Integral(FI)
The Fuzzy Integral (FI) approach
evaluates the performances of different processors
by a fuzzy measurement. It is a meaningful
formalism for combining classifier outputs that can
capture interactions among the various sources of
information. A popular fuzzy integral approach
Sugeno Integral method is used.
However there are some detection and
recognition problems from the above three methods
(Spectral Change Difference algorithm, Feature and
decision level fusion techniques).So Curvelet
Transform is being used to overcome these
drawbacks.
1.4 Curvelet Transform based image fusion:
Most natural images/signals exhibit line-
like edges, i.e., discontinuities across curves (so-
called line or curve singularities). The fusion of
multi-temporal satellite images is a very useful
technique in various applications of remote sensing.
Since edges play a fundamental role in image
representation, one effective means to enhance
spatial resolution is to enhance the edges. The
curvelet-based image fusion method [10] provides
richer information in the spatial and spectral
domains simultaneously. Curvelet Transform is a
new multi-scale representation most suitable for
objects with curves. It was developed by Candes
and Donoho (1999).
Consider the images at two different dates.
Curvelet Transform is applied to these two images.
The exact reason for choosing the curvelet
algorithm is because of its sparsity. Therefore the
curvelet construction is now based on the
consideration of polar coordinates in frequency
domain and also to construct curvelet elements
being locally supported near wedges. Curvelets are
designed to handle curves using only a small
number of coefficients. Hence it handles curve
discontinuities well. The Curvelet Transform
includes four stages:
1.4.1 Sub-band decomposition
It divides the image into resolution layers. Where
each layer contains details of different frequencies:
such as Low-pass filter and Band-pass (high-pass)
filters.
1.4.2 Smooth partitioning
A grid of dyadic squares is defined. Here
windowing function is applied
1.4.3 Renormalization
Renormalization is centering each dyadic square to
the unit square [0,1]x[0,1] and each square is
renormalized. The renormalized ridges has an
aspect ratio of widthlength
2
.For encoding these
ridges efficiently, ridgelet transform is used.
1.4.4 Ridgelet analysis
Each normalized square is analyzed in the ridgelet
system.
III. EXPERIMENTS AND RESULTS
3.1 Experimental data
This section describes the study area, the
urban changing event and the available remote-
sensing images for testing and evaluating the
proposed automated change detection algorithm.
India's mega-cities and its 4,000 cities and
towns account for 60% of the gross domestic
product. In 2001 the percentage of urban population
is 28%. 35 cities have population more than a
million, compared to 23 cities in 1991. By 2015
more than half of Indians are projected to be urban
dwellers, 1/3 will be slum dwellers and squatters. In
the national capital region 1/2 of households and
1/3 of population were migrants and a little less
than 1/2 of migrants were from rural areas. The
urban to urban stream of migration was found to be
more important for larger cities like Delhi. To this
aim, Landsat Multispectral Scanner (MSS) and
Landsat Thematic Mapper (TM) images are used as
experimental data source for urban expansion
monitoring and land cover change detection task in
the urban areas of New Delhi. In Fig. 2. The red
color denotes the urban areas.
(a) (b)
Fig. 2. Remotely sensed images of New Delhi (a)
May 8, 1974; (b) April 21, 1999
S.Anita Johannah Golda, B.Bala Murugan / International Journal of Engineering Research
and Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 2, March -April 2013, pp.1546-1551
1550 | P a g e
3.2 Results:
(a) (b)
(c) (d)
(e) (f)
(g) (h)
(i)
Fig. 3. Change detection results (a)Y
SD
; (b) Y
SR
; (c)
Y
AD
;(e) Y
ED
;(f)Y
CST
;(h)FS feature level
fusion;(i)MV decision level fusion;(j)DS decision
level fusion;(k)FI decision level fusion.
The performance is evaluated in terms of
overall accuracy, kappa statistic, omission and
commission. Overall accuracy is the percentage of
pixels correctly classified. Kappa statistic accounts
for number of pixels correctly classified by chance
and it is always lower than overall accuracy. The z-
test is defined as:
2
1
2
2
1 2
k k
k k
z
o o
=
(7)
Where K
1
and K
2
are two selected kappa
coefficients
2
k1
and
2
k2
are variances. Omission is
the pixels that are placed in another class which are
omitted. Commission is the pixels placed in a class
in which they do not belong.
TABLE I
Comparison of Fusion Scheme with Curvelet Based
Fusion Scheme for Change Detection
Fig. 4. Comparison graph of accuracy values for
different approaches
0
20
40
60
80
100
120
Y
s
d
Y
s
r
Y
a
d
Y
e
d
Y
c
s
t
F
S
M
V
D
S
F
I
A
c
c
u
r
a
c
y
Approaches
Analysis of accuracy values
Fusion scheme
change detection
Curvelet and
fusion scheme
based change
detection
Fusion
level
Fusion
method
Fusion scheme for change detection
Overall
accuracy
(%)
Kappa
Omission
(%)
Commission
(%)
SCD
Y
SD
78.5678 0.7213 18.4224 3.0098
Y
SR
80.0882 0.4140 19.5654 0.34641
Y
AD
78.2377 0.7289 18.1446 3.6176
Y
ED
69.5033 0.8316 4.5752 25.9216
Y
CST
78.2149 0.7210 18.1111 3.674
Feature
level
FS 70.4118 0.8279 4.9306 24.6577
Decision
level
MV 78.2639 0.7267 18.1364 3.5997
DS 69.4583 0.8317 4.567 25.9747
FI 81.3954 0.8242 4.1528 14.4518
Curvelet based fusion scheme for change detection
SCD
Y
SD
78.5931 0.6640 18.6185 2.7884
Y
SR
80.3007 0.0833 19.6895 0.0098039
Y
AD
72.0605 0.4489 18.7304 9.2092
Y
ED
81.6887 0.8180 3.5605 14.7508
Y
CST
80.9534 0.8316 3.4788 15.5678
Feature
level
FS 97.7263 0.7924 0.60703 1.6667
Decision
level
MV 79.71 0.6325 18.6961 1.594
DS 97.0842 0.8586 0.33007 2.5858
FI 98.3317 0.8265 0.42484 1.2435
S.Anita Johannah Golda, B.Bala Murugan / International Journal of Engineering Research
and Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 2, March -April 2013, pp.1546-1551
1551 | P a g e
IV. CONCLUSION
The change detection method based on
fusion of multiple Spectral Change Difference
(SCD) data set is feasible and more effective for
urban expansion monitoring and land cover change
detection over urban areas than the change
information extracted by single SCD algorithms.
From the experimental results in New Delhi study
area, it is confirmed that the limitation and
uncertainties caused by a single SCD image can be
reduced using suitable fusion techniques. Fusion
methods combine the various SCD images. Feature
level fusion can effectively reduce omission errors
and Decision level fusion is good at restraining
commission errors, but both of them lead to an
increase of the overall accuracy and the optimum
fused image can be obtained using Curvelet
Transform.
In future, the challenging problem such as
computational cost of curvelet transform can be
overcome by exploring the suitable thresholding
function that incorporate and exploit the special
characteristics of the curvelet transform.
ACKNOWLEDGEMENT
The authors would like to thank land
remote sensing satellite for providing Landsat MSS
and Landsat TM datasets.
REFERENCES
[1] Peijun Du, Sicong Liu, Paolo Gamba,Kun
Tan, and Junshi Xia, Fusion of
Difference Images for Change Detection
Over Urban Areas, IEEE Journal of
selected topics in Applied Earth
Observations And Remote Sensing, Vol.
5, No. 4, August 2012.
[2] H. Nemmour and Y. Chibani, Multiple
support vector machines for land cover
change detection: An application for
mapping urban extensions, ISPRS J.
Photogramm. Remote Sens., vol. 61, no. 2,
pp. 125133, 2006.
[3] L. Castellana, A. DAddabbo, and G.
Pasquariello, A composed supervised/
unsupervised approach to improve change
detection from remote sensing, Pattern
Recogn. Lett., vol. 28, no. 4, pp. 405413,
2007.
[4] C. E. Woodcock, S. A.Macomber, M. Pax-
Lenney and W. B. Cohen, Monitoring
large areas for forest change using
Landsat: Generalization across space, time
and Landsat sensors, Remote Sens.
Environ., vol. 78, no. 12, pp. 194203,
2001.
[5] T. Sohl, Change analysis in the United
Arab Emirates: An investigation of
techniques, Photogramm. Eng. Remote
Sens., vol. 65, no. 4, pp. 475484, 1999.
[6] T. Celik, Unsupervised change detection
in satellite images using principal
component analysis and k-means
clustering, IEEE Geosci. Remote Sens.
Lett., vol. 6, no. 4, pp. 772776, 2009.
[7] F. Bovolo and L. Bruzzone, A theoretical
framework for unsupervised change
detection based on change vector analysis
in the polar domain,
[8] P. Gong, Change detection using
principal component analysis and fuzzy
set theory, Can. J. Remote Sens., vol. 19,
no. 1, pp. 2229, 1993
[9] Y. Lu, P. F. Shi, and Y. M. Zhao, Voting
principle for combination of multiple
classifiers, J. Shanghai JiaoTong
University, vol. 34, no. 5, pp. 680683,
2000.
[10] Jianwei Ma and Gerlind Plonka, The
Curvelet Transform, IEEE signal
processing magazine, March 2010
[11] Peter Rogers and Sumeeta Srinivasan
(1999) Land Use Change in Developing
Countries Comparing India and China
DEAS/HUCE, Harvard University,
Cambridge
[12] LANDSAT TM and LANDSAT MSS
images are obtained from the online
source
http://www.mdafederal.com/environment-
gis/national-security.../land-use.