Image Fusion An Application
Image Fusion An Application
ISSN 2229-5518
Abstract- In today’s era digital image processing has wide areas of application. One of the most
important applications of image processing is in image fusion. Image fusion is a technique which is used
for combining relevant information from two or more images into a single image. This fused image
contains more information than any of the two input images. The fused image can have complimentary
spatial and spectral resolution characteristics. We can apply this method in remote sensing application as
IJSER
well as satellite imaging application. In this proposed work, two images are fused based on the wavelet
transform using different fusion technique. The objective of this Proposed work is to fuse two images in
such a way that we can get better result which contain more information. The respected results will
evaluate using parameters like Pseudo Signal to Noise Ratio, Root Mean Square Error, Standard
Deviation and Entropy.
Keywords— Image Fusion, Wavelet Transform, Peak Signal to Noise Ratio, Root Mean Square Error,
Standard deviation, Entropy, Spatial resolution, Spectral resolution.
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1248
ISSN 2229-5518
Figure 1.1 Basic Image Fusion Scheme Sweta Mehta, et. al. [3] introduced the
Curvelet Transform and uses it to fuse
Above figure shows the basic image
images. The experiments show that the
fusion scheme. In this figure image 1 and
method could extract useful information
image 2 are combined to extract the
from source images to fused images so that
information between those two images into a
clear images are obtained. In this paper we
single image. This figure consist of the two
put forward an image fusion algorithm based
section namely that is redundant information
on Wavelet Transform and the Curvelet
and complimentary information. The
Transform. Low and high frequency
IJSER
reliability of image is improved by
coefficients are choosen according to
redundant information and the capability of
different frequency domain after Wavelet
image is improved by complimentary
and the Curvelet Transform. In choosing the
information.
low frequency coefficients, the concept of
local area variance was chosen to measuring
2. Related Work
criteria. In choosing the high frequency
Various types of method are proposed in
coefficients, the window property and local
literature review which are used for image
characteristics of pixels were analyzed.
fusion.
Finally, the proposed algorithm in this
Navneet kaur, et. al [1] Presented
article was applied to experiments of multi-
“Review On: Image Fusion Using Wavelet
focus image fusion and complementary
and Curvelet Transform” Image fusion
image fusion.
refers to the process of combining the
information from two or more images into a
Yufeng Zheng, et al.[4] Presented The
single highly informative image. The
fusion performance of the advanced DWT
resulting fused image contains more
(aDWT) method proposed here was
information than the input images. In this
compared with six other common methods,
paper, two medical images are fused based
and, based on the four quantitative
on the Wavelet Transform (WT) and
measures, was found to perform the best
Curvelet transform using different fusion
when tested on the four input image types.
techniques. The objective of the fusion of an
Since the different image sources used here
MR image and CT image of the same organ
varied with respect to intensity, contrast,
is to obtain a single image containing as
noise, and intrinsic characteristics, the
much information as possible about that
aDWT is a promising image fusion
organ for diagnosis.
procedure for inhomogeneous imagery.
Myungjin Choi, et al. [2] introduced a
new method based on a curvelet transform,
which represents edges better than wavelets.
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1249
ISSN 2229-5518
IJSER
image (MRI), scan provides different types after applying the fusion rule fused image is
of information, by fusing them we can get extracted. The proposed work is given
accurate information for better clinical below.
diagnosis.
The algorithm proposed by the authors 1. To study the different techniques for
has its own pros and cons, with respect to image fusion.
that we will going to propose efficient 2. To study different transforms with
architecture for image fusion which will be respect to image fusion in digital image
used in various domain, also the statistical processing.
parameters like peak signal to noise ratio,
entropy, standard deviation and root mean 3. To design image fusion techniques
square error will be evaluated to prove our using transform method.
method will be efficient with respect to
others proposed method. 4. Verification of results based on
design techniques
3. System Architecture
Image Fusion is a method where we 3.1 Introduction to Wavelet Transform
add or merge two images to acquire useful
Information. It is very useful technique. For Wavelet transform is one of the most
efficient tool for image processing. In this
Fusing Images we can use different
transform Techniques. Figure below shows proposed work two types of wavelet
the block diagram for wavelet transform transform are used.
based image fusion. 3.1.1 Discrete Wavelet Transform.
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1250
ISSN 2229-5518
IJSER
is applied to get the fused image. rule to those images. After that inverse
wavelet transform is applied to get the fused
The discrete wavelet transform (DWT) is an image.
implementation of the wavelet transform in The stationary wavelet transform is very
which discrete set of the wavelet scales and much similar to discrete wavelet transform
translations obeying some defined rules. In only difference is that the process of down
other words, this transform decomposes the sampling is supressed that’s why stationary
signal into mutually orthogonal set of wavelet transform is translation invariant. It
wavelets, which is the main difference from does so by suppressing the down-sampling
the continuous wavelet transform (CWT), or step of the decimated algorithm and instead
its implementation for the discrete time up-sampling the filters by inserting zeros
series sometimes called discrete time between the filter coefficients. Algorithms in
continuous wavelet transform(DTCWT). which the filter is up-sampled are called “à
trous”, meaning “with holes”. As with the
decimated algorithm, the filters are applied
first to the rows and then to the columns. In
this case, however, although the four images
produced (one approximation and three
Figure 3.1.1.2 Block diagram of DWT and detail images) are at half the resolution of
IDWT the original; they are the same size as the
original image. The 2-D SWT
Above block diagram shows the decomposition scheme is illustrated in
reconstruction process of inverse discrete Figure 4.3.2.1[13].
wavelet transform from discrete wavelet
transform. Where h(n), g(n), h1(n) and g1(n)
can be constructed by using quadrature
mirror filter(QMF)[12].
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1251
ISSN 2229-5518
IJSER
the resultant data[14].
3.2.1 Peak Signal to Noise Ratio(PSNR)
The mathematical relation for PSNR is
given below, 3.2.3 Entropy
Entropy is defined as amount of
information contained in a signal. Shannon
PSNR is defined as log of the ratio between was the first person to introduce entropy to
the square of the peak value to the Mean quantify the information. The entropy of the
Square Error multiplied to the value 10. This image can be evaluated as,
basically projects the ratio of the highest
possible value of the data to the error
obtained in the data. 5 In our case, at pixel
level, the highest possible value is 255. i.e. Where G is the number of possible gray
in a 8 bit gray scale image, the maximum levels, P(di) is probability of occurrence of a
possible value is having every bit as 1 –> particular gray level di . Entropy can directly
11111111; which is equal to 255. Then the reflect the average information content of an
error between the fused image and the input image. The maximum value of entropy can
image is calculated as the Mean Square be produced when each gray level of the
Error and the ratio value is obtained. If both whole range has the same frequency. If
the fused and the input images are identical, entropy of fused image is higher than parent
then the MSE value would be 0. In that case, image then it indicates that the fused image
the PSNR value will remain undefined[14]. contains more information[14].
3.2.4 Standard Deviation
3.2.2 Root Mean Square Error
To find RMSE we have to first calculate This metric is more efficient in the absence
Mean Square Error(MSE) which is given by of noise. It measures the contrast in the
the following relation,
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1252
ISSN 2229-5518
fused image. An image with high contrast As shown in above flow chart the first
would have a high standard deviation. step is to take the first two images from the
image database. The next step is to apply the
wavelet transform to those images. In this
work there is two type of wavelet transform
are studied that is discrete wavelet transform
and stationary wavelet transform. After
Where hI f (i) is the normalized histogram of application of wavelet transform wavelet
the fused image If (x,y) and L is number of coefficient are generated. After that fusion
frequency bins in histogram[15]. rule is applied to those coefficients. The next
step is to apply inverse wavelet Transform to
those coefficient to extract the original
4. Result and Discussion coefficient. At last the fused image is
4.1 Introduction obtained.
In this section result and discussion based
on discrete wavelet transform and stationary
wavelet transform are carried out. The
flowchart of proposed work is given below. 4.2 Result Based On Discrete Wavelet
IJSER
Transform-This shows the result based on
Discrete wavelet transform and the
performance parameters like peak signal to
noise ratio(PSNR), root mean square
error(RMSE), standard deviation(SD) and
entropy(E) are evaluated.
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1253
ISSN 2229-5518
IJSER
Transform Table 4.3.1 Image Fusion Based on
Stationary Wavelet Transform
In this section the result based on
Stationary wavelet transform and the 4.4 Result Analysis
performance parameters like peak signal to
4.4.1 Result Analysis based on Discrete
noise ratio(PSNR), root mean square Wavelet Transform.
error(RMSE), standard deviation(SD) and
entropy(E) are evaluated. Image Discrete Wavelet Transform
Name PSNR MS RMS SD Entrop
Input Input Fused E E y
Image 1 Image 2 Image Image 37.51 132. 11.52 53.87 0.006
1 5 77 2 8
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1254
ISSN 2229-5518
4.4.2 Result Analysis based on Stationary transform is more as compare to the value of
wavelet Transform. standard deviation by stationary wavelet
transform is less. The value of entropy by
Imag Stationary Wavelet Transform discrete wavelet and the value of entropy by
e PSNR MSE RMS SD Entrop
Name E y stationary wavelet transform is near about
Imag 38.17 97.81 9.88 52.39 0 same. So from above discussion for overall
e1 8 14 9 2 tested images stationary wavelet transform is
Imag 41.36 22.56 4.75 46.05 7.060 best.
e 3 9 0 4
REFERENCES
2
[1] Navneet Kaur, Madhu Bahl, Harsimran
Imag 38.76 74.60 8.67 52.17 7.352 Kaur, “review on : Image Fusion Using
e 6 7 3 4 Wavelet and Curvelet Transform”, (IJCSIT)
3 International Journal of Computer Science
Imag 43.05 10.36 3.21 42.19 0.000 and Information Technologies, Vol. 5 (2) ,
e 3 0 8 8 9 2014, 2467-2470.
4 [2] Myungjin Choi, Rae Young Kim,
Imag 39.80 46.16 6.94 29.18 6.444 Myeong-Ryong NAM, and Hong Oh kim,
e 9 0 1 8 “The Curvelet Transform For Image
IJSER
Fusion”.
[3] Sweta Mehta, Prof. Bijith Marakarkandy,
“CT AND MRI Image Fusion Using
Table 4.4.2.1 Evaluation of Result Based on Curevelet Transform” Issn: 0975 – 6779|
Stationary Wavelet Transform. Nov 12 To Oct 13 | Volume – 02, Issue – 02.
[4] Yufeng Zheng, Edward A. Essock and
CONCLUSION Bruce C. Hansen Dept. of Psychological &
In this proposed work image fusion Brain Sciences, University of Louisville,
method is introduced which is based on the Louisville, “An Advanced Image Fusion
wavelet transform, which incorporated the Algorithm Based on Wavelet Transform –
statistical parameters like peak signal to Incorporation with PCA and Morphological
noise ratio(PSNR), root mean square Processing”.
error(RMSE), standard deviation(SD) and [5] Chavez Jr, P. S., Sides, S. C. &
entropy are evaluated to prove our method Anderson, J. A. 1991, “Comparison of three
will be efficient with respect to others different methods to merge multiresolution
proposed method. As shown in above result and multispectral data: Landsat TM and
for overall images the value of PSNR by SPOT
discrete wavelet transform is less as panchromatic. Photogrammetric Engineering
compare to the value of PSNR by stationary and Remote Sensing” 57 (3) 295–303.
wavelet transform is more. The value of [6] Aiazzi, B., Alparone, L., Baronti, S. &
MSE by discrete wavelet transform is more Garzelli, A. 2002. Context-driven fusion of
as compare to the value of MSE by high spatial and spectral resolution data
stationary wavelet transform is less. based on oversampled multi resolution
Although the value of RSME by discrete analysis. IEEE
wavelet transform is more as compare to the Transactions on Geoscience and Remote
value of RMSE by stationary wavelet Sensing 40 (10) 2300–2312.
transform is less. Although the value of [7] Alparone, L., Baronti, S., Garzelli, A. &
standard deviation by discrete wavelet Nencini, F. 2004a. A global quality
IJSER © 2015
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 11, November-2015 1255
ISSN 2229-5518
IJSER
Implementation,” Photogrammetric
Engineering and Remote Sensing, vol. 66,
2000, pp. 49-61.
[11] L. Wald, T. Ranchin and M. Mangolini,
“Fusion of Satellite images of different
spatial resolution: Assessing the quality of
resulting images”,Photogrammetric
Engineering and Remote Sensing, vol. 63,
no. 6, 1997, pp. 699.
[12] Shih-Gu Huang (黃世谷),“Wavelet for
Image Fusion”, Graduate Institute of
Communication Engineering & Department
of Electrical Engineering, National Taiwan
University.
[13] Mirajkar Pradnya P., Sachin D. Ruikar
“Image Fusion Based On Stationary Wavelet
Transform” Electronics &
Telecommunication Dept. Sinhgad Acadmy
of Engg. Pune.
[14] Dr. Pankaj Mohindru, Maninder Singh
Gill, Dr. Pooja “A New Image Fusion
Algorithm Based on,Wavelet Transform and
Adaptive Neuro Fuzzy Logic Approach”
International Journal of Advanced Research
in Electrical, Electronics and
Instrumentation Engineering Vol. 3, Issue 8,
August.
IJSER © 2015
http://www.ijser.org