IRJET- Lunar Image Fusion based on DT-CWT, Curvelet Transform and NSCT

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 08 | Aug 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 766
Lunar Image Fusion Based on DT-CWT, Curvelet Transform and NSCT
Tulasi Bharathi1, K. Jhansi Rani2
1PG Scholar, Dept. of ECE, UCEK JNTUK, Kakinada, A.P, India.
2Assistant Professor, Dept. of ECE, UCEK JNTUK, Kakinada, A.P, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The process of combining information from
high spectral resolution lunar image and high spatial
resolution lunar image is known as multi sensor lunar
image fusion. The multi spectral lunar image usually
consists of images in different bands. The features and
missing locations of various moon surface images from
various instruments can be extracted and viewed accurately
using image fusion. Previously various methods have been
used for multi sensor image fusion. To extract features of the
moon surface, this paper involves image fusion methods
based on Dual Tree-Complex Wavelet Transform (DT-CWT),
Curvelet transform and Nonsubsampled Contourlet
transform (NSCT). The spectral resolution of the fused
image is preserved after image data fusion. The fused lunar
image also shows good spatial information. From the results
of statistical evaluation parameters demonstrated for the
study site, the results of the used techniques are found and
are compared.
Key Words: Curvelet Transform, DTCWT, Image Fusion,
Lunar Image, Multi Sensor, NSCT.
1. INTRODUCTION
The process of combining two or more images to form a
new image by using a certain fusion rule is known as
image fusion [1][2]. The use of image fusion in current
image processing systems is increasing, mainly because of
the sophisticated image acquisition techniques. There are
many conventional image fusion algorithms. These can be
classified as (a) projection and substitution methods, as
Intensity- Hue-Saturation (IHS) color fusion, and Principal
Component Analysis (PCA) fusion; (b) band ratio and
arithmetic combination, such as Brovey, multiplicative
and Synthetic Variable Ratio (SVR), and (c) the wavelet
fusion techniques in which spatial features from high
resolution image is fused into low resolution image [7][9].
Image fusion methods can be classified into two –
Transform domain and spatial domain. The fusion
methods such as averaging, Brovey method, PCA and IHS
based methods come under spatial domain fusion. In case
of spatial domain methods, there is a drawback. There will
be spectral distortion in the fused image. Spectral
distortion can be reduced by using transform domain
image fusion. The multi resolution analysis has become a
very useful tool for analyzing remote sensed images
[4].The wavelet transform concentrate more on vertical
and horizontal features. In order to retain features from
different angles, the curvelet transform is useful [3][5].
But NSCT is multi scale in nature. With its flexibility, multi
direction quality, and shift invariant image decomposition,
it can be efficiently implemented in image fusion [8]. This
method shows a better performance in spatial and
spectral quality of the fused image compared to other
methods of fusion.
In this paper, NSCT fusion method is proposed to
obtain appreciable spatial and spectral resolution and the
results are compared with DTCWT and Curvelet transform
methods. This paper is organized as follows. In Section II
describes the study area and dataset used, Section III
describes the methodology for DTCWT, curvelet transform
and NSCT, Section VI discusses the results and analysis of
the fused image and Section V concludes the paper with
future work.
2. IMAGE DATASET
Lunar image data set is acquired from the Indian satellite
Chandrayaan-1, HySI and TMC. Hyperspectral Imager
(HySI) of Chandrayaan-1 has spatial resolution of 80 m and
performed mapping in the 400-900 nm band with spectral
resolution of 15nm with swath coverage of 20 km. Terrain
Mapping Camera (TMC) of Chandrayan-1 has spatial
resolution of 5 m with 20 km swath. The input study area is
shown in Fig -4. (a) and (b).
3. METHODOLOGY
3.1DT-CWT Fusion
The DT-CWT has its real and imaginary parts of transform
split into two trees. The real part of this complex wavelet
is obtained as the difference of two separable wavelets
and is oriented in -450. Real 2D wavelet oriented at +450.
Four more oriented real wavelets in the direction of +750,
-750, +150 and -150 can be obtained. The real part of the
complex Dual Tree wavelet Transform alone constitutes
the Real Oriented 2D Dual Tree Transform. This transform
has six distinct directions and there will be two wavelets
in each direction. It gives a perfect reconstruction as the
filters are chosen from a perfect reconstruction bi-
orthogonal set. It is applied to lunar images using complex
filtering which is separable, in two dimensions. It can be
chosen over DWT as it gives good results. The
methodology of DTCWT based image fusion is shown in
Fig -1.

The low spatial resolution multispectral image is a
histogram matched with the high resolution image. The
complex wavelet transform is applied to high spatial
resolution image and low spatial resolution image. In
complex wavelet transform two DWT with different filters
are used, this decomposes the image into two
approximations subbands and six detailed subbands.
Fig -1: Architecture for Dual Tree Complex Wavelet
Transform (DTCWT) Image fusion
The two approximation subbands has the low
frequency components which are filtered using two
different opposite direction filter. The six detailed
subbands have high frequency components, which are
symmetrically opposite in direction. The approximation
subbands are fused using averaging fusion rule and the
detailed coefficients are also fused using the averaging
method. Then the inverse complex wavelet transform is
applied to get the fused image.
3.2 Curvelet Fusion
Curvelet is used to represent an image at different scales
and different angles. It is like an extension of the wavelet
transforms [6][11]. The curvelet transform is the multi
scale transform which consists of directional elements.
The methodology of curvelet image fusion is shown in Fig -
2.
Fig -2: Architecture for Curvelet Image Fusion
The low spatial resolution multispectral image is
resampled to the size, high spatial resolution image using
a bilinear interpolation technique. Curvelet transform is
performed to both high spatial resolution image and the
low spatial resolution image; both the images are
decomposed into the approximation and detailed
coefficients. The low frequency components of the
decomposed image are fused using the averaging method.
The high frequency components are fused using different
fusion rules like maximum method. Inverse curvelet
transform to obtain the fused image.
3.3 NSCT Fusion
NSCT is a flexible and shift invariant computational
framework suitable for lunar images. There will be two
stages, Non-Subsampled Directional filter bank and Non-
Subsampled Pyramid [10]. In the non-subsampled
directional filter bank, the direction decomposition occurs
in x stages in high frequency from non-subsampled
pyramid at each scale and gives 2x directional subimages
as input image. With the help of two channel filter banks,
one image of high frequency and one image of low
frequency can be produced at each level of non-
subsampled pyramid decomposition. The subsequent
decomposition is performed to decompose the low-
frequency components of the image. Non-Subsampled
Pyramid results in y+1 subimages. It consists of one low-
frequency image and y high-frequency images with same
size as that of the input image, where y is number of levels
of decomposition.

Fig -3: Architecture of NSCT Image Fusion
Firstly, we take the image and apply NSCT to get low-
frequency and high-frequency subimages. Then fusion of
low-frequency subimages is done using Pulse coupled
neural network (PCNN) method. Later the fusion of high-
frequency sub-image is also done using PCNN method.
After this, l-level Inverse NSCT is performed on the fused
low-frequency and high-frequency sub-images is done to
get the fused image. The process of NSCT image fusion is
shown in Fig -3.
4. RESULTS AND DISCUSSIONS
Fusion of the high spatial resolution image of TMC and
high spectral resolution image of HySI is performed using
the three methodologies stated in Section III and the
output of study area is shown in Figure 5.
Using visual and statistical parameters, the fused
images are compared. It is observed that most of the fused
images shows good enhancement in spatial details of the
moon surface due to different illumination angle of the
input images. Visual comparison indicates that NSCT
fusion shows better visual interpretability of craters
compared to the other fusion techniques. The contrast
features present in high resolution image and low
resolution image are highly visible in the fused image. But
the spectral characteristics of the multispectral image
were well preserved in NSCT fusion since the fusion takes
in the spectral domain, which is evident from Correlation
coefficient, Entropy and Average gradient in Table 1.
(a)
(b)
Fig -4: (a) High spectral resolution image (Chandrayaan
HySI 80m), (b) High spatial resolution image
(Chandrayaan TMC 5m)
(a)
(b)

(c)
Fig -5: (a) DTCWT Fusion, (b) Curvelet Fusion and
(c) NSCT Fusion
Table -1: Statistical Evaluation of Image Fusion for the
input Study Area
Statistical
Parameter
Band
Fusion Techniques
DTCWT Curvelet NSCT
Correlation
Coefficient
Band 1 0.9920 0.9928 0.9987
Band 2 0.9975 0.9982 0.9994
Band 3 0.9863 0.9868 0.9976
Band 4 0.9819 0.9825 0.9967
Band 5 0.9983 0.9995 0.9996
Standard
Deviation
Band 1 41.3228 41.3370 47.1759
Band 2 38.9459 38.9778 41.7462
Band 3 38.7510 38.7860 41.8890
Band 4 38.1223 38.1933 41.0190
Band 5 36.5783 36.5970 36.9006
Entropy
Band 1 0.3960 0.3279 0.4604
Band 2 0.3889 0.3218 0.5512
Band 3 0.3883 0.3218 0.3466
Band 4 0.3835 0.3218 0.2867
Band 5 0.3873 0.3218 0.5360
RMS Error
Band 1 21.8023 20.8726 2.3813
Band 2 12.6555 11.6747 1.4963
Band 3 11.1648 10.4636 2.9502
Band 4 9.2809 8.8848 3.3769
Band 5 3.0423 1.7691 1.0671
Average
Gradient
Band 1 18.0173 19.8021 28.3858
Band 2 15.8016 16.7597 21.9935
Band 3 14.9826 16.0375 21.3210
Band 4 14.2921 15.4590 20.0218
Band 5 14.3247 14.7708 17.0609
Band 5 14.3247 14.7708 17.0609
The correlation coefficient gives the relation between
the original image and the fused image. Its range is from 0
to 1. If the correlation coefficient is 0, this means that
there is no relation between the fused image and the
original image. If the value is 1, it means that the fused
image is highly related to the original image. Standard
deviation means the deviation of the fused image from the
original image. Entropy is used to calculate the amount of
information in an image. High entropy means high amount
of information is being retained. RMS Error represents the
amount of error between the fused image and the original
image. The low value of RMSE represents better
performance. Average Gradient describes the amount of
visual information present in an image.
From the results of correlation coefficient, the method
of NSCT fusion retains 99.84 % of spectral information.
The method of DTCWT retains 99.12 % and the curvelet
method retains 99.19 %. This implies that the NSCT image
fusion retains more spectral information than that of the
DTCWT and curvelet image fusion. The standard deviation
of the NSCT fusion is 41.74614 whereas DTCWT and
curvelet fusion have the values of 38.74 and 38.77
respectively. The entropy of the NSCT fused image is
0.4362 which is more when compared to DTCWT and
curvelet fused image entropy values of 0.3888 and 0.3230
respectively. From RMS error, it is revealed that the NSCT
fused image has less error compared to the DTCWT and
curvelet fused images. The average gradient value of the
NSCT fusion method is 21.76, which is more than the value
of DTCWT and curvelet methods having 15.48 and 16.57
respectively. Overall, the NSCT fused lunar image has good
clarity and better performance when compared to fused
lunar images of the other techniques used.
5. CONCLUSION
The average values of statistical parameters considered
are calculated for the fusion methods to obtain the
statistics. The obtained statistics are compared with the
dual tree complex wavelet transform and curvelet
transform. The higher value in the average gradient and
entropy indicates that the spatial information is much
retained in the NSCT method. Higher values in correlation
coefficient indicate that the spectral information is well
preserved in the NSCT fusion. The lower value of root
mean square error of NSCT method shows that there is
less amount of error between the fused and the original
image. Good spatial fidelity is shown by NSCT image
fusion. The spectral features are also retained in a good
manner in case of NSCT fusion method. So, it can be
concluded that nonsubsampled contourlet transform
(NSCT) produced better results for the lunar image data
fusion compared to the other methods used.
ACKNOWLEDGMENT
We acknowledge the use of data from the Chandrayaan-1,
first lunar mission of the Indian Space Research
Organization (ISRO), archived at the Indian Space Science
Data Center (ISSDC).

REFERENCES
[1] A. A. Goshtasby and S. Nikolov, “Image fusion:
Advances in the state of the art,” Information Fusion,
vol. 8, no. 2, pp. 114–118, Apr. 2007.
[2] Pohl et al. “Multi-sensor image fusion in remote
sensing: concepts, methods and applications”,
International Journal of Remote sensing, Vol. 19, No.5,
pp. 823-854, 1998.
[3] B. Y. Shutao Li, “Multifocus image fusion by combining
curvelet and wavelet transform,” Pattern Recognition
Letters, ELSEVIER, no. 28, p. 1295-1301,
2008.
[4] D. Hall and J. Llinas, "An introduction to multisensor
data fusion", Proceedings IEEE, Vol. 85(1), pp. 6-23,
1997.
[5] LucStarck, E J., Donoho, D L., “The curvelet transform
for image denoising”, IEEE Transactions on Image
Processing, vol.11, pp:670-684, 2002.
[6] Ali, FE., Dokany, IM., Saad, AA., Samie, FEA., “Fusion of
MR and CT images using the curvelet transform”, pp. 1
– 8, 2008.
[7] Gaurav Choudhary, Dr. Anvesh Verma, “A Review on
digital image fusion techniques”, International Journal
of Advancement in Engineering and Technology,
Management and Applied Science (IJAETMAS), Vol. 4,
Issue 6, June 2017, pp. 88-102.
[8] Kurakula Sravya, Dr. P. Govardhan, Naresh Goud M,
“Image fusion on multi focused images using NSCT”,
International Journal of Computer Science and
Information Technologies, Vol 5(4), 2014, 5393-5396.
[9] Dr. S. Narayana Reddy, Y. S. Vaishnavi, “Performance
comparison of different image fusion techniques for
multi-focus and multi-modal images”, International
Journal of Science and Research (IJSR), Volume 5,
Issue 12, December 2016.
[10] V. Bhateja, A. Srivastava, A. Moin and A. Lay-Ekuakille,
"NSCT based multispectral medical image fusion
model," 2016 IEEE International Symposium on
Medical Measurements and Applications (MeMeA),
Benevento, 2016, pp. 1-5.
[11] A. A. Micheal and K. Vani, "Multi-sensor image fusion
of the lunar image data using DT-CWT and curvelet
transform," 2017 4th International Conference on
Electronics and Communication Systems (ICECS),
Coimbatore, 2017, pp. 49-53.
[12]Qu Xiao-Bo, YAN Jing-Wen, XIAO Hong-Zhi, ZHU Zi-
Qian. Image Fusion Algorithm Based on Spatial
Frequency-Motivated Pulse Coupled Neural Networks
in Nonsubsampled Contourlet Transform Domain.
Acta Automatica Sinica, Vol.34, No.12, pp: 1508-
1514.Dec.2008.

IRJET- Lunar Image Fusion based on DT-CWT, Curvelet Transform and NSCT

More Related Content

What's hot (20)

Similar to IRJET- Lunar Image Fusion based on DT-CWT, Curvelet Transform and NSCT (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Lunar Image Fusion based on DT-CWT, Curvelet Transform and NSCT