SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Performance Analyzing of High Resolution Pan-Sharpening

Techniques: Increasing Image Quality For Classification
Using Supervised Kernel Support Vector Machine

Yuhendra, H.Kuze, and J. T. Sri Sumantyo

Center for Environmental Remote Sensing
Chiba University
1-33, Yayoi, Inage, Chiba 263-8522, Japan
E-mail:Yuhendra@graduate.chiba-u.jp

Abstract : - Pan-sharpening is also known as image fusion, resolution merge, image integration, and
multi sensor data fusion has been widely applied to imaging sensors. The purpose of pan-sharpening
is to fuse a low spatial resolution multispectral image with a higher resolution panchromatic image to
produces an image with higher spectral and spatial resolution. In this paper, we investigated these
existing pan-sharpening methods based on visual and spectral analysis. And to achieve assess the
accurate classification process, we proposed a support vector machine (SVM) based on radial basis
function (RBF) kernel. In the Experimental results, a comparative performance analysis of
techniques by various methods show that Gram-Schmidt followed by PCA perform best among all
the techniques. Besides that, higher overall accuracy of Gram-Smidth (GS) fused image increase 0.90
percent. And also, the high producer’s and user’s accuracy average of Gram-Smidth (GS) fused for
each of the classes and methods used was always reported greater than 91.8% and 91.11%,
respectively, indicating the overall success of the performed classification. And the followed by PCA
was 90.84% and 89.99.
Keywords: Pan-sharpening, multispectral, resolution, classification, support vector machine(SVM),
radial basis function (RBF).

1 Introduction Orbview, etc. It have one bands of spectral

In recent years, many satellite imaging system
were orbited, the availability and various of high
resolution from remote sensing data satellite image
from space has increased. And nowadays,
Multisensor data fusion has become a discipline
to which more and more general formal solutions to
a number of application cases are demanded[1]. He
has become very important issue in many
applications of remote sensing like for visual
interpretation, object identification, land use
classification, for detecting changes, for updating
maps, for monitoring hazards, visualization purposes
and for many other applications.
The availability and variations data collected
from airborne sensors have resulted in various data
sets of different spatial, spectral and temporal
resolutions. The high spatial resolution data
satellite, such as IKONOS, SPOT-5, Quickbird,
information in panchromatic (Pan) image and four
bands in multispectral (MS) image (i.e band
1(blue), band 2 (green), band 3(red), and band 4
(near infrared). In order to produces panchromatic
image can be used to enhance the resolution of
multispectral image through a process known as
sensor fusion or pan-sharpening. Pan-sharpening is
shorthand for panchromatic sharpening. It means
using a panchromatic (single band) image to
increase the spatial resolution of the multispectral
image. Pan-sharpening is also known as image
fusion[2], resolution merge, image integration, and
multi sensor data fusion [3]. And image pan-
sharpening is a pixel level fusion technique that
describes a process of changing a set of low (coarse)
spatial resolution multispectral images to high (fine)
spatial resolution color images, by fusing a geo
registered fine spatial resolution black/white

ISSN: 1792-5088 260 ISBN: 978-960-474-233-2

 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

(panchromatic) image of the same area. The problem based on spectral quality analysis. And to
panchromatic image is usually obtained from the achieve accurate classification a spectral image, has
same platform and taken at the same time or in a be done tested the classification on the pan-
very short time duration with the multispectral sharpened images using kernel support vector
image. machine (radial basis function) Classification. In
Many researchers have been addressed the this paper, we applied intensity-hue-saturation
problem of multiresolution image fusion for (IHS), CN-Brovey, principal component analysis
remote-sensing applications and pan-sharpening (PCA), Gram-Schmidt, and CN-Spectral methods.
methods have been developed [4]-[6]. Diversity of
proposed pan-sharpening method have been 2 Remote Sensing Data Set
developed in a high-resolution Pan image with an The Quikbird satellite, launched on 18 October
multi spectral remote-sensing image such as based 2001 in Vanderberg Air Force Base, California,
on transform methods including the intensity-hue- USA is a high resolution satellite operated by
saturation (IHS) [7]–[9], Brovey [10],[11], principal Digital Globe. The Quickbird satellite data
component analysis (PCA) [12],[13]. And other consisted of one band panchromatic image at 0.61-
method Multi resolution analysis based on intensity m resolution and four band multispectral images at
modulation such as Gram-Schmidt [14], CN- 2.44-m multispectral and a 61 cm panchromatic
Spectral [15], High Pass Fusion [16],[17]. Several resolution image at nadir, and a 2.88 m
researcher have proposed using the wavelet multispectral and a 72 cm panchromatic image. The
transform to extract geometric edge information specifications of spectral wavelengths and spatial
from the panchromatic image [18]-[21], and resolutions for Quickbird imagery are listed in
Discrete wavelet transform [22]. Table 1.
In image classification process, support vector Table 1
machines (SVM) have grown speed up and shown a Spectral Pixel
great potential as classifiers for remotely sensed Sensor Range Band Scene Size Resolution
450-520 nm 1(B)
data in recent years. Support vector machines 520-600 nm 2(G) 2.44-2.88
(SVM) non-parametric classifiers have the 16.5 x 16.5 (m)
Multispectal 630-690 nm 3(R)
(km)
additional advantage that they are able to 760-900 nm 4(NIR)
Panchromatic 450-900 nm 1 61-72 cm
simultaneously minimize the empirical
classification error and maximize the class
separation using various transformations of hyper Figure 1 show the spectral response curve for the
planes [23]. This allows SVM to better deal with difference bands of the satellite imagery. It shows
data of high dimensionality and classes with a that the sensor has low spectral response in the
feature space of multimodal distribution producing band 1(blue), and maximum in the band 2 (green)
often better results in comparison to other and band 3 (red). It also shows that though the
parametric or non-parametric methods[24]-[27]. spectral range of the panchromtic sensors are
SVM have been addressed successfully in many provided as 0.45 – 0.90 m.
fields [28]-[32]. Recently, they have been
successfully used in remote sensing area [25]-[27]
and [33]-[36].
The main approach of Pan-sharpening is how
to integrating the spatial of the high resolution
panchromatic (Pan) image and the spectral
information of low resolution multispectral image
(MS) to produce a high resolution MS image. In
this paper, our motivation is to generate high
resolution panchromatic image (Pan) and
multispectral images (MS) with more information
and better quality image based on some methods Fig. 1. QuickBird Spectral Response
process and considered as a statistical estimation

ISSN: 1792-5088 261 ISBN: 978-960-474-233-2

 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING
2.1 Quickbird Image
Panchromatic imagery from Quickbird has a spatial
resolution of about 0.7 meters at nadir. The same
sensor collects the nearly the multispectral data at
about 2.8 meters resolution. The sensors can
collect co registered (explained below)
panchromatic and four-band (red, green, blue and
near-infrared) multispectral images. In this study
part of the Quickbird image of the Sundarbans
(Vidyasagar Bridge) area, the India, obtained in
November 2006 is used. Figure 2 show the subset
of QuickBird panchromatic and multispectral
image.
A B
Fig.2. panchromatic (A), multispectral image (B)
3 Pan-sharpening Methods
In the images processing, we have implemented six
pan-sharpening methods to analysis and performed
the image fusion.
Intensity, Hue &Saturation: The IHS fusion
technique is widely used in image fusion to exploit
the complementary nature of multispectral images
[7]-[9]. Three bands of a multispectral image are
considered as in color image. The HIS fusion
method converts a color image from the red, green,
and blue (RGB) space into the IHS color space.
Then the intensity band (I) in the IHS space is
replaced by a high-resolution panchromatic image
and then transformed back into the original RGB
space together with the previous hue band (H) and
the saturation band (S), resulting in an IHS fused
image.
Colour Normalized-Brovey: The Brovey methods
provide good spatial quality but poor spectral
quality. Use of Color normalized-Brovey
sharpening to apply a sharpening method that uses
a mathematical combination of the color image and
high resolution data. The gray level ( GL) values
for each band in a set of multispectral images are
divided according to the sum of gray levels (GL) in
the number (n) of bands used in the process. The
result is then multiplied by the gray level (GL) of
ISSN: 1792-5088
the band with the best spatial resolution, in this
case, the panchromatic (Pan) band [10],[11].
Gram-Schmidt (GS): The GS spectral sharpening
method [14] enhances the spatial resolution of the
multispectral (MS) image by merging the high
resolution panchromatic (Pan) image with the low
spatial resolution. In the simulated high resolution
panchromatic (Pan) band is employed as the first
band. Then the high resolution panchromatic (Pan)
band is swapped with the first GS band. Finally, the
inverse GS sharpening transform is applied to form
the pan-sharpened spectral bands. The low
resolution multispectral bands are merged into a
single, then simulated low resolution panchromatic
(Pan) by calculating the appropriate weight.
The Principle Component Analysis (PCA):The
principal component analysis method for pan-
sharpening abbreviated as PCA is the other known
method of image pan-sharpening that belongs to the
projection and substitution method categories.
Principal component analysis can easily be
understood as the transformation of correlated
variables into a set of uncorrelated variables called
principal components. PCA is applied to the
original image . The first principle component
(PC) image is replaced by the panchromatic
image [12]. In the PCA-based method, the PCA
transform converts intercorrelated multispectral
bands into a new set of uncorrelated components. It
is assumed that the first PC image with the
highest variance contains the most amount of
information from the original image and will be the
ideal choice to replace the high spatial resolution
panchromatic image [13].
CN Spectral Sharpening : CN Spectral Sharpening
is stands for of the Color Normalized algorithm
often used to pan-sharpening multispectral (RGB)
images. CN Spectral can be used to
simultaneously sharpen any number of bands and
retain the input image's original data type and
dynamic range. Input multispectral bands are
sharpened only if they fall within the spectral range
of one of the pan-sharpening image's bands. The
spectral range of the panchromatic is defined by
the band center wavelength and full width-half
maximum (FWHM) value, both obtained from the
panchromatic image. MS bands are grouped into
spectral segments defined by the spectral range of
262 ISBN: 978-960-474-233-2
 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING
the panchromatic band. Each multispectral band is
multiplied by the panchromatic band, then
normalized [15].
3.1 Pan-sharpening Assessment
Pan-sharpening Assessment based on quantitative
and statistical analysis obtained from the
experimental result. The Quantitative analysis used
to visual and spectral quality of fused images. In
this study, the spectral quality of the fused image
can be measured by the correlation coefficient
(CC), Standard deviation (SD), mean, maximum
value, and covariance.
4. Kernel SVM Classifier
Support vector machines (SVM) are supervised
learning algorithms based on statistical learning
theory, which are considered as heuristic
ntrepresented by a feature vector. Set of training
example data, the SVM classifier calculates an
optimal hyper plane characterized by a vector that
provides the best separation between the two
classes. The optimal hyper plane is the one that
maximizes the distance between the hyper plane
and the nearest positive and negative training
example, called the margin. In order to represent
more complex shapes thanlinear hyper planes, the
classifier may use kernel functions which replace
the vector. In this case, the problem transforms into
an equivalent linear hyper plane problem of higher
(sometimes infinite) dimensionality. Commonly
used SVM kernels include linier, polynomial ,
gaussian also known as radial basis function (RBF),
sigmoid and normalized kernels.
Linear : K
,  ) =  . 
1
p
Polynomial : K
,  

 .  
 (2)
algorithms[28]. SVM is a binary classification
method that provides a separation of classes by
fitting an optimal separating hyper plane to a set of
training data that maximizes the separation between
the classes. An optimum hyper plane is determined
using train data sets and its generalization ability is
verified using test data sets. For the classification
of k dimensional data sets, k - 1 dimensional hyper
plane is produced with SVM. It can seen from Fig.
3a, there exist various hyper planes separating two
classes. However, there is only one hyper plane that
provides maximum margin between the two classes
(Fig. 3b), which is called the optimum hyper plane
[28]. The points that constrain the width of the
margin are called support vectors [37], [38].
Fig. 3. Hyper planes for linearly separable
data (a) Optimum hyper plane and support
vectors (b).
Gaussian : K
,  ) =exp (- .  .   (3)
Sigmoid : K
,  ) =tanh ( .  .  + a)
(4) Normalized : K’
,  ) = 
 .  ) K
,
(5)  ) ½ K ( .  )1/2
where  and  represent a set of training data and
is a user-defined parameter, specified as the inverse
of the number of the spectral bands of the sensor on
which the SVM is implemented. The penalty
parameter can be introduced to the SVM classifier
to quantify the misclassification error, providing
important information in the case of non-separable
training datasets.
The selection of the SVM kernel classifier is
regarded as the most important task in the
implementation and the performance of the SVM
classifier [39],[40]. To classification of Pan-
sharpened Image, we was used the radial basis
function (RBF) kernel. Since it has been
widely used in the literature in land classification
and change detection studies with
various satellite data [39-44] has shown
classification results that have outperformed
those obtained by other kernel types. In addition,
RBF kernel requires only one parameter to be
pre-defined, which makes it more robust in its
implementation in contrast to other kernels
[43],[45].

The success of the SVM depends on how well the 5. Result And Analysis
process is trained. Each training example is 5.1 Visual Assessment

ISSN: 1792-5088 263 ISBN: 978-960-474-233-2

 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Visual Figure 4 (A) and (B) is pair of Quickbird Table 3 Spectral assessment result
original Image. We can show that panchromatic Methods Color Recovery Sharpness
(Pan) image has better spatial resolution than the HIS Not acceptable Very Good
multispectral (MS) image (low resolution). All CN-Brovey Not acceptable Very Good
fused image seems to be very close to the pan PCA Good Good
image in terms of spatial quality. A vegetation Gram-
when seen in the near infrared is very bright (higher Schmidt
Very Good Very Good

reflectance). So, the vegetation pixels come out CN-

much to bright. When intensity image is replaced Spectral
Not acceptable Very Good

by pan image, the vegetation pixels are much too

bright result in poor color recovery. We can be
concluded that the color recovery IHS fused image
(C) is better but still not perfect, spatial in
produces very sharp image, spatial information and
sharpness is good. The CN-Brovey fused image (D)
can also see that the color provides very good
contrast but some color are distorted (blurriness).
The color recovery PCA fused image (E) result is
much better than (C), (D) and (G) result but still
have some miss registrations around the bright
object. PCA algorithm is dependent on scene
content, for Instance high vegetation content causes
poor performance. High NIR contribution tends to
distort the PCA transformation and cause
blurriness.
A B C exception, where the mean, standard deviation and
5.2 Spectral Assessment
In Table 3 it is possible to verify that even though
good result were obtained by pan-sharpening
methods using Gram-Schmidt, the PCA methods
outperformed all the others. This methods presented
fused value that were very close to those of the
Quickbird original bands for practically all the
bands under this study. PCA and Gram-Schmidt
were closest to the mean, standard deviation ,
covariance , and coefficient correlation values of
the Quickbird original in band 1, 2 and 3. For the
mean values, in band 2, values obtained from
Gram-Schmidt method were same with the
Quickbird original bands. Band 1 was only the
covariance obtained from PCA methods was
closest to Quickbird original bands.

Table 4 Spectral assessment result

G
Methods Mean STD Cov. Corr.
Band 1
QuickBird 229 16 270 1.000
HIS 89 58 3388 1.000
CN-B 80 25 645 1.000
D E F
PCA 229 17 294 1.000
GS 229 15 227 1.000
CNS 232 33 1091 1.000
Band 2
QuickBird 296 34 542 0.956
HIS 99 66 3572 0.929
Fig. 3.(A) Original MS image, (B) Original Pan image, (C) CN-B 88 30 481 0.613
HIS Fused image, (D) CN-Brovey Fused image, (E) PCA PCA 292 37 622 0.963
Fused image, (F) Gram-Schmidt Fused image and (G) CN GS 296 31 449 0.944
Spectral Fused image CNS 301 54 1747 0.973
Band 3
The result of the Gram-Schmidt fused image (F) is QuickBird 171 36 571 0.945

quite similar to PCA . The color recovery is better HIS 83 53 2842 0.919
CN-B 78 21 161 0.291
and image are sharper. Based on the visual
PCA 171 38 641 0.951
assessment we can get the summary of the
GS 171 33 474 0.933
comparison (Table 2) for each of methods. CNS 174 45 1320 0.877

ISSN: 1792-5088 264 ISBN: 978-960-474-233-2

 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Bands STD Values

Based on the spectral assessment result, we can
plotted the statistical parameters in show Figure 4- 80
7. 60
40
20
Bands Mean Values 0

400
300
200
100 Band 1 Band 2 Band 3
0

Fig. 7. Standard deviation of the original & fused images

Table 5 Classification accuracy

Band 1 Band 2 Band 3 Class Methods (Fused image)
HIS Brovey GS PCA CNS
Overall Accuracy (%)
Fig. 4 Mean of the original & fused images 86.78 83.74 90.77 89.59 87.73
Kappa Coefficient
0.83 0.79 0.88 0.87 0.84
Bands Coeff.Cor Values Producers Accuracy (%)
Sea 91.40 88.94 98.56 97.92 99.60
1 Soil 91.32 92.38 92.50 92.33 87.41
0,8 Vegetation 95.39 95.02 99.12 98.05 99.58
0,6 Shadow 75.83 72.66 87.37 83.83 88.67
0,4 Building/Roof 95.59 93.84 96.77 97.00 74.32
0,2 Bridge/Road 69.79 51.27 76.48 75.92 76.20
0
Average 86.55 82.35 91.8 90.84 87.63
User Accuracy (%)
Sea 85.61 82.29 91.96 89.82 97.27
Soil 87.22 85.49 88.76 88.62 77.61
Vegetation 95.25 95.06 97.94 95.88 98.13
Shadow 86.02 77.34 91.61 90.73 92.32
Band 1 Band 2 Band 3 Building/Roof 88.95 89.93 95.62 97.07 85.20
Bridge/Road 70.08 61.52 80.77 77.82 81.37
Average 85.52 81.93 91.11 89.99 88.65
Fig. 5.Correlation between original & fused images

Bands Covariance Values 5.2 Accuracy Assessment

The classification accuracy was evaluated using the
4000 confusion matrix use to show the accuracy of a
3000
classification result by comparing a classification
2000
1000
result with using ground truth region of interests
0 (ROIs). In each case, an overall accuracy, kappa
coefficient, producer and user accuracies are reported
(Table 5).
To achieve accurate classification a spectral image,
has be done tested using supervised kernel support
Band 1 Band 2 Band 3
vector machine (radial basis function). As seen, the
higher overall accuracy of result using the Gram-
Fig. 6. Covariance of the original & fused images Smidth (GS) fused image was 0.90 percent, PCA,
CNS, HIS and Brovey fused image up to 0.80 percent
respectively. Whereas the Kappa coefficient result
greater than 0.83, exceptional Brovey is 0.79. Also,
the high producer’s and user’s accuracy average of

ISSN: 1792-5088 265 ISBN: 978-960-474-233-2

 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING
Gram-Smidth (GS) fused the classes and methods
used was always reported greater than 91.8% and
91.11%, respectively, indicating the overall success of
the performed classification. And the followed by
PCA was 90.84% and 89.99.
6. Conclusion
We have implemented six pan-sharpening methods to
analysis and performed the image fusion. The visual
and spectral analysis of the fused images indicates the
results generated with Gram-Schmidt and followed
PCA methods preserves spectral and spatial
information of the objects in original images better
than the results generated by IHS, Brovey and CN-
Spectral methods. The statistical comparison of this
methods shows that the Gram-Schmidt and PCA
fused image has the best spectral , too. Whereas in
the classification process, we are propose to use
supervised kernel function (radial basis) support
vector machines (VCM). In the experimental result
that kernel radial basis function (RBF) is work out
the proof of the accuracies, kappa, producers, and
users performances.
References:
[ 1 ] G.Andrea , N. Filippo, Panchromatic sharpening
of remote sensing images using
multiscaleKalman filter, the journal of the pattern
recognition society, 40 3568–3577,2007.
[ 2 ] U. Kumar, C. Mukhopadhyay, T.V.
Ramachandra, Pixel based fusion using
IKONOS imagery, International Journal of
Recent Trend in Engineering Vol1, No.1, May
2009.
[ 3 ] V. Vijayaraj, N.H. Younan, C.G O’Hara,
Concepts of image fusion in remote sensing
application, IEEE Trans. Geosciences Remote
sensing, vol 10 no.6, pp. 3781-3784, June 2006.
[ 5 ] A. Garzelli, F. Nencini, L. Alparone, B. Aiazzi,
and S. Baronti, Pansharpening of multispectral
image: A critical review and comparison,” in
Proc. IGARSS, 2004, vol.1 pp. 81–84.
[ 6 ] C. Pohl and J. L. Van Genderen, Multisensor
image fusion in remote sensing: Concepts,
methods, and applications, Int. J. Remote Sens.,
vol. 19, no. 5, pp. 823– 854, 1998.
[7] M.Choi, H.C.Kim, N.Cho, and H. O.Kim, An
Improved Intensity-Hue-Saturation Method for
IKONOS Image Fusion, submitted to IJRS,2000.
[ 8 ] M.Choi, A New Intensity-Hue-Saturation Fusion
ISSN: 1792-5088
Approach to Image Fusion with a tradeoff
Parameter, IEEE Trans. Of Geosci. And Remote
Sens., vol.44, pp. 1672-1682, 2006.
[ 9 ] T.-M. Tu, P. S. Huang, C.-L. Hung, and C.-P.
Chang, A Fast Intensity-Hue-Saturation Fusion
Technique With Spectral Adjustment for IKONOS
Imagery, IEEE Geosci. and Remote Sens. Lett.,
vol. 1, no. 4, October 2004.
[10] Q. Du, N. Younan, R. King, and V. Shah, On the
performance evaluation of pan-sharpening
techniques, IEEE Geosci. and Remote Sens. Lett.,
vol. 4, no. 4, pp. 518–522, October 2007.
[11] Pohl, C., Van Genderen, J.L., Multisensor image
fusion in remote sensing: concepts, methods and
applications, International Journal of Remote
Sensing 19, 823-854, 1998.
[12] V. K. Shettigara, A generalized component
substitution technique for spatial enhancement of
multispectral images using a higher resolution
data set, Photogramm. Eng. Remote Sens., vol.
58, no. 5, pp. 561–567, 1992
[13] Shah, Vijay P., Nicolas H. Younan, and Roger L.
King. An Effictient PanSharpening Method Via a
Combined Adaptive PCA Approach and
Contourlets. IEEE Transaction on Geoscience
and Remote Sensing 46 (2008).
[14] C. A. Laben, and B. V. Brower, Process for
enhancing the spatial resolution of multispectral
a
imagery using pan-sharpening, United States
Eastman Kodak Company (Rochester, NY). US
Patent 6011875, 2000. URL:
http://www.freepatentsonline.com/ 6011875.html.
[15] J. Vrabel, P. Doraiswamy, J. McMurtrey and A.
Stern, Demonstration of the Accuracy of
Improved Resolution Hyperspectral Imagery,
SPIE Symposium Proceedings, 2002.
[16] P. S. Chavez, Jr., S. C. Sides, and J. A. Anderson,
Comparison of three different methods to merge
multiresolution and multispectral data: Landsat
TM and SPOT panchromatic, Photogramm. Eng.
Remote Sens., vol. 57, no. 3, pp. 295–303, 2001.
[17] L. Wald, T. Ranchin, and M. Mangolini, Fusion
of satellite images of different spatial resolutions:
Assessing the quality of resulting images,
Photogramm. Eng. Remote Sens., vol. 63, no. 6,
pp. 691–699, 1997.
[18] E. Yu and R. Wang,Fusion and enhancement of
the multispectral image with wavelet transform,
Comput. Eng. Sci., vol. 23, no. 1, pp. 47–50,
2001.
[19] J. Zhou, D. L. Civico, and J. A. Silander, A
266 ISBN: 978-960-474-233-2
 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING
wavelet transform method to merge Landsat TM
and SPOT panchromatic data, Int. J. of Remote
Sens., vol. 19, no. 4, 1998
[20] R. King and J. Wang, A wavelet based algorithm
for pan sharpening landsat 7 imagery, Geosci.
and Remote Sens. Symp., 2001. IGARSS ’01.
IEEE 2001 Int., vol. 2, pp. 849–851, 2001.
[21] N. Memarsadeghi, J. Le Moigne, and D. Mount,
Image fusion using cokriging, Geosci. and
Remote Sens. Symp., 2006. IGARSS 2006. IEEE
Int. Conf. on, pp. 2518–2521, August 2006.
[22] Z. Li, Z. Jing, X. Yang, S. Sun, Color transfer
based remote sensing image fusion using non-
separable wavelet frame transform, Pattern
Recognition Lett. 26 (13) (2005) 2006–2014.
[23] G. P. Petropoulos,W. Knorr1, M. Scholze1, L.
Boschetti, G. Karantounias, Combining ASTER
multispectral imagery analysis and support
vector machines for rapid and cost-effective post-
fire assessment: a case study from the Greek
wildland fires of 2007, Nat. Hazards Earth Syst.
Sci., 10, 305–317, 2010
[24] Waske, B. and Benediktsson, J.A.: Fusion of
support vector machines for classification of
multisensor data, IEEE T. Geosci. Remote Se.,
45(12), 3858–3866, 2007.
[25] Sanchez-Hernandez, C., Boyd, D. S., and Foody,
G. M.: Mapping specific habitats from remotely
sensed imagery: Support vector machine and
support vector data description based
classification of coastal saltmarsh habitats, Ecol.
Inform., 2, 83-88, 2007.
[26] Van der Linden, S., Janz, A., Waske, B., Eiden,
M., and Hostert, P.: Classifying segmented
hyperspectral data from a heterogeneous urban
environment using support vector machines, J.
Appl. Remote Sens., 1, 013543,
doi:10.1117/1.2813466, 2007.
[27] Z. Wang and A. C. Bovik, A universal image
quality index, IEEE Signal Process Lett., vol. 9,
no. 3, pp. 81 – 84, March 2002.
[28] T. Kavzoglu , I. Colkesen, A kernel functions
analysis for support vector machines for
land cover classification, International Journal of
Applied Earth Observation and Geoinformation,
11, 352–359, 2009.
[29] Chi Hau Chen, Pei-Gee Peter Ho, Statistical
pattern recognition in remote sensing, Pattern
Recognition 41, 2731 – 2741, 2008.
[30] B. Nahla , B. Andrew, Rule extraction from
support vector machines: A review,International
Journal of Neurocomputing, Pebuary, 2010.
[31] K. Milos, B. Branislav, G. Bosko, Soil type
classification and estimation of soil properties
using support vector machines, International
ISSN: 1792-5088
Journal of Geoderma 154, 340-347, 2010
[32] Lau, K.W., Wu, Q.H., Local prediction of non-
linear time series using support vector
regression. Pattern Recognition 41, 1556–1564,
2008.
[33] J.R. Otuke i,T. Blaschke, Land cover change
assessment using decision trees, support vector
machines and maximum likelihood classification
algorithms, International Journal of Applied
Earth Observation and Geoinformation 12S,
(S27–S3, 2010.
[34] Lihong Su, Optimizing support vector machine
learning for semi-arid vegetation mapping
by using clustering analysis, ISPRS Journal of
Photogrammetry and Re mote Sensing 64,
407-413, 2009.
[35] K. Jan, R. Andreas, R. Volker, K. Tobias, K.
Jacek, H. Patrick, Land cover mapping of large
areas using chain classification of neighboring
Landsat satellite images, Remote Sensing of
Environment 113, 957–964, 2009.
[36] B. Wu, Z. G. Xiong,Y. Z Chen, Y. Zhao,
Classification of quickbird image with maximal
mutual information feature selection and support
vector machine, Procedia Earth and Planetary
Science 1,1165–1172, 2009.
[37] Sherrod, P.H., Classification and regression trees
and support vector machines for predictive
modeling and forecasting,
http://www.dtreg.com/DTREG.pdf., 2003
[38] Oommen, T., Misra, D., Twarakavi, N.K.C.,
Prakash, A., Sahoo, B., Bandopadhyay, S., An
objective analysis of support vector machine
based classification for remote sensing,
Mathematical Geosciences 40, 409–424,2008.
[39] Keuchel, J., Naumann, S., Heiler, M., and
Siegmund, A.: Automatic land cover analysis for
Tenerife by supervised classification using 2003.
[40] Li, D.-C. and Liu, C.-W.: A class possibility based
kernel to increase classification accuracy for
small data sets using support vector machines,
Expert Syst. Appl., 37, 3104–3110, 2010.
[41] Carrao, H., Concalves, P., and Caetano, M..:
Contribution of multispectral and multitemporal
information from MODIS images to land cover
classification, Remote Sens. Environ., 112, 986–
997, 2008.
[42] Knorn, J., Rabe, A., Radeloff, V. C., Kuemmerle,
T., Kozak, J., and Horstert, P.: Land cover
mapping of large areas using chain classification
of neighboring Landsat satellite images, Remote
Sens. Environ., 113, 957–964, 2009.
[43] Huang, C., Davis, L.S., Townshed, J.R.G., An
assessment of support vector machines for land
cover classification, International Journal of
267 ISBN: 978-960-474-233-2
 SELECTED TOPICS in POWER SYSTEMS and REMOTE SENSING

Remote Sensing 23, 725–749,2002.

[44] Kavzoglu, T., Reis, S., Performance analysis of
maximum likelihood and artificial neural network
classifiers for training sets with mixed pixels,
GIScience and Remote Sensing 45, 330–342,
2008
[45] Huang, C., Song, K., Kim, S., Townshend, J. R.
G., Davis, P., Masek, J. G., and Goward, S. N.:
Use of dark object concept and support vector
machines to automate forest cover change
analysis, Remote Sens. Environ., 112, 970–985,
2008.

ISSN: 1792-5088 268 ISBN: 978-960-474-233-2