IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

Performance Analysis of Wavelet Functions in

Fusion of MRI and CT Images
Renjith V. Ravi∗ , Sujith M.V.† , Shafeen.K.M‡ , Thamjid Ali Asharaf U§ , Sajidh C.T.¶ , Sayooj Mohan M.Tk
Department of Electronics and Communication Engineering, MEA Engineering College,
Malappuram, Kerala, India.
Email : ∗ [email protected], † [email protected], ‡ [email protected],
§ [email protected], ¶ [email protected], k [email protected]

Abstract—The fusion of images is the mechanism by which in transform domain approach Laplacian, Curvelet, Counterlet,
two or more images are merged into one image with important Wavelet [5]. This work is done in the wavelet domain as it
features. Fusion is an important technology in many different ar- helps us to view different frequency components of image in
eas, including remote sensing, robotics and medical applications.
The image fusion results in a composite image that is ideally very easy and useful method. Fusion in terms of frequency can
suited for human and machine perception or external image be said as fusion of similar frequency components together.
processing tasks. In medical imaging technology, the Magnetic Although wavelet transformation has been widely used [7],
Resonance Image (MRI) highlights the soft tissue of the body [8], [9] for the fusion of MRI and CT images, wavelets such as
and Computed Tomography (CT) provides a better view on Symlets and Coiflets have not been found to be used. Most of
hard tissue highlighting bones so their fusion will lead to better
information content. Highlight of this particular image fusion the works were based solely on Daubechies or biorthogoanal
is, one of the most useful diagnoses of tumor it provides the type wavelets. So, in this article, we carried out a complete
identification of gross tumor volume and clinical target volume analysis of most of the types of wavelet transformations.At
by 80% more comparing to the MRI and CT images can provide any rate, more than one wavelet from each family could also
by itself. In this paper,we compared the efficiency of different be used for analysis. Also, in most of the previous works,
fusion techniques. The wavelet based image fusion techniques
comprises of two steps among which the first step is Discrete there has been no study on how the result varies as the level
Wavelet Transform (DWT) based decomposition of two input of decomposition. The evaluation metrics considered in the
images into four coefficients each such as approximation, vertical, other works were only one or two which had a chance to vary
horizontal and diagonal and fusion of each respective coefficients depending on the input image. An analysis based on three
is performed based on some particular fusion rules. the fusion metrics was therefore found to be necessary.
rules can be used for this particular application are Maximum,
Minimum and Mean.Various parameters like Entropy, Mutual In this article, we analysed the performance all types of
Information and Standard Deviations were used to evaluate the wavelet families with level of decomposition from 1 to 5,
fused image. in fusion of MRI and CT of brain images. Along with this,
Index Terms—Image Fusion, wavelet based image fusion, performance of various fusion rules along with various wavelet
Fusion Rules, Fusion Techniques, Fusion Performance families at each level of decomposition also evaluated. Section
II discuss about materials and methods used for this research,
I. I NTRODUCTION section III shows the results and its discussion, and finally the
Image Fusion is one of the major fields of study in image article is concluded in section IV.
analysis. It is a means of merging valuable information from
different images into a single resultant image, which is more II. M ATERIALS AND M ETHODS
precise and complete than the input image data. Image fusion A. Discrete Wavelet Transform
is a good way to obtain the most important functionality for DWT [10] [11] is a method by which images are de-
certain applications. [1], [2], [3]. Image fusion techniques were composed to provide a set of coefficients that represent the
created to produce a new image, which is more relevant to image as a set of subbands with different frequencies. It is
human vision or perception of the system. The Main advantage a well-known method of signal analysis. The 2-D DWT can
of image fusion is to improve reliability and capability. transform an image from a spatial domain to a wavelet domain
There are different methods available for image fusion once it is decomposed.During a single decomposition, the
in which the most popular one is pixel level fusion [4]. It image is divided into four coefficients such as Approximation,
is further classified into four types, they are substitution, Horizontal, Vertical and Diagonal.
mathematical, optimization and transform domain [5]. The
substitution technique has been divided into PCA, Averaging, B. Rules for Image Fusion in Wavelet Domain
IHS, and color mixed. The mathematical fusion [6] is based on For the fusion of the images obtained after wavelet decom-
brovey transform, the optimization approach is defined under position, we can choose either of different fusion rules or its
Bayesian and neural networking, and there are four methods suitable combination [12], [13], [14]. Let M and N are two

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.
 IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

images of the same size representing the MRI and CT images. Step 2:Use different fusion rules to fuse each corresponding
Then the first rule of fusion is the maximum to obtain the fused frequency components (For Eg. Approximation co-
image F is shown in Eq.1. efficients of MRI image will be fused with Approx-
imation coefficients of CT image. Horizontal coeffi-
F = (M × Q) + (N × ∼ Q) (1)
cients of MRI image will be fused with Horizontal
Where Q is absolute value of M and N . The Second rule is coefficients of CT image and so on).
minimum in which fused image F is shown in Eq. 2. Step 3:Carry Inverse DWT (Use same levels of decompo-
sition) on the fused components to reconstruct the
F = (M × Q) + (N × ∼ Q) (2) fused image.
Third fusion rule is mean as shown in Eq.3. For more clarification and reproduction of the work, the
flowchart of experimental setup is shown in Fig.2 and Pseudo
F = (M + N )/2 (3) code is shown in Algorithm 1 .
The fourth fusion rule used is random is same as shown in
Eq.1.But here Q is a Boolean random matrix.
MRI Image CT Image
C. Experimental Setup
The experimental set up for performance analysis of image
fusion using various wavelet bases and fusion rule is shown in Obain the Size of Image
Fig.1. The figure illustrates only a single level decomposition.

MRI Image CT Image

Wavelet Decomposition Wavelet Decomposition

Check
No
Whether
Resize
Approximation Horizontal Vertical Diagonal Approximation Horizontal Vertical Diagonal
the Size
are equal

Fusion Rules Fusion Rules Fusion Rules Fusion Rules

Yes
Fused Approximation Fused Horizontal Fused Vertical Fused Diagonal
Coefficients Coefficients Coefficients Coefficients

Reconstruction Using Inverse Wavelet Transform Decompose the Image using DWT

Fused Image

Fig. 1. Experimental Setup for Image Fusion using DWT

Apply Fusion Rule
But, for the experiment, we have conducted decomposition
upto five levels. Now the wavelet decomposition has been done
using various wavelet functions. The family of wavelets used
Reconstruct the Image using IDWT
in this particular experiment are Haar, Daubechies, Coiflet,
and Symlets, Biorthogonal, Reverse Biorthogonal, Discrete
Mayer etc. From these wavelet family about 54 sub wavelets
were used and each levels of the wavelet decomposition have Fused Image
been done for different pair of MRI and CT images. Then
different kind of fusion rules (as mentioned in Subsection
II-B) was also tried to find out the best fusion method among
Evaluation Metrics
maximum, minimum, random and linear techniques. Each of
these results were evaluated using different evaluation metrics
Fig. 2. Flowchart of experimental setup
like Entropy, Mutual Information, Standard deviationSD and
the best fusion technique was found.
The experimental process is simply explained below in step III. R ESULTS AND D ISCUSSION
by step manner; The performance of the methods discussed above was
Step 1:Apply DWT to both the input images for doing the further evaluated by using experiments in order to validate our
first level decomposition (Appy DWT to the low research. A total of 10 patients were studied with a focus on
frequency components for further level of decom- visualization of brain structures using CT and MRI data vol-
position, if any.). ume sets and considered one set for the experiment after con-

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.
 IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

Algorithm 1: Pseudo code for Wavelet Based Image

Fusion
Input: MRI Image,CT Image,Level of DWT
Decomposition,Fusion Rules
Output: Fused Image (F),Evaluation Metrics
1 Find the Size of Both Images
2 Check whether their size are equal,
(I) If Size are Equal, then go to Step 3.
(II) If Size are not equal, then resize and go to Step 3.
3 Take DWT of Both MRI and CT image upto the Level
of DWT Decomposition for obtaining the Coefficients,
4 Both MRI and CT images corresponding decomposed
coefficients will be fused using fusion rule.
5 Take inverse DWT to reconstruct the fused image.
6 Metrics are evaluated to find the best fusion algorithm

sulting with medical doctors. In order to find the best fusion Fig. 3. MRI Image for Fusion
algorithm, we have undergone a detailed analysis of all wavelet
transformations, fusion rules and parameters. We considered
eight wavelet families ’dmely’, ’Haar’, ’Daubechies(db)’,
’Symlets(sym)’, ’Coiflets (coif)’, ’biorthogonal(bior)’,’reverse
biorthogonal(rbior)’ and ’discreet mayor(dmey)’ for the fusion
of CT and MRI medical images.Each pair of images has been
analyzed with more than 50 transformations, i.e. we have taken
every level of decomposition of each wavelet transform for
analysis. Here we’ve looked at the analysis of one CT and
MRI image pair.
The three considerations to be made while doing the per-
formance analysis of wavelet based image fusion are; (i)
Type of wavelet function to be used, (ii) Levels of wavelet
decomposition, and (iii) fusion rule which is applied.
The input MRI and CT images are shown in Fig.3 and Fig.
4 respectively. The fused images using (min-min) (ie minimum
for both approximation and detailed coefficients) is shown
in Fig. 5. correspondingly (mean-mean) and (max-max) are
shown in Fig.6 and Fig.7 respectively.
We have used three non-reference fusion metrics for ob-
jective assessment of fusion performance: Mutual Information Fig. 4. CT Image for Fusion
(MI) [15] [16], Entropy(E) [17], [18], [8] and standard devia-
tion (SD) [17], [8]. The experiment is conducted for this pair of
images for all the available wavelet families for decomposition daubche’s, biorthogonal, reverse biorthoganal and discrete
levels from 1 to 5 for all the available fusion rules.As the mayor were used for the analysis. Different fusion rules were
quantity of results received were too much, we have included tested, including the ’mean’ rule, ’maximum’ rule, ’minimum’
only best results in this article. These results are shown in rule and ’random’ rule. Here ’maximum’ rule with 5 levels of
Table I. decomposition produces better result, so it is selected. Finally,
Further, to analyze the performance of image fusion accord- we find out the best fusion algorithm or the best fusion method
ing to the decomposition level, we have taken bior 3.1 as it for the biomedical image fusion. From the table of results,
shows better performance than other with (max-max) rule. The ’biorthogonal3.1’ wavelet indicates the best results showing
results are shown in Fig.8,Fig.9,Fig.10,Fig.11 and Fig.12 and good values for evaluation metrics.
the values are available in Table II. For the analysis we have considered the important param-
eters like entropy, standard deviation and mutual information.
A. Discussion of Results After the literature survey it is found out that all the discussed
The wavelet based fusion with separate wavelet transforma- parameters high enough for a better-quality fused image. From
tions and fusion rules was tested on human brain CT and MRI the above Table I, it is clearly seen that fusion undergone with
images. Different wavelet transforms like haar, coiflet, symlet, biorthogonal wavelets show high values for the metrics. When

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.
 IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

Fig. 7. Fused Image using (max,max) Rules

Fig. 5. Fused Image using (min-min) Rules

Fig. 6. Fused Image using (mean,mean) Rules

Fig. 8. Fused Image at Level of decomposition 1

analyzing the values, ’biorthogonal3.1’ has got the highest

value. And also it is noted from the results especially in Table have taken 3 pairs of MRI and CT images and have undergone
II that, as the level of decomposition increases, the values of a fusion analysis by considering all the types and rules of
metrics are also increasing. ie for the maximum level, there fusion, and thus we can see that biorthogonal wavelets are
is a gradual improvement or variation in the values of all the better than others because there is only limited variation in
metrics. the values of their parameters.For the analysis, we have taken
For each type of fusion, the maximum fusion rule has more than 500 results, including all types of fusion methods
the maximum result, so that we can see that the maximum- and fusion rules, for more than two sets of images, and
maximum (max-max) fusion rule is the best fusion rule that finally, after the overall analysis, we find that biorthogonal3.1
can be adopted for brain image fusion. Neither the mean nor wavelet is the best fusion method for biomedical image fusion
the minimum are good values for the fused images. Another techniques. This wavelet family can be used for any fusion
important wavelet, which shows good parametric values, is technique, especially in the medical field, for better-quality
Daubechies (db), especially the db10 of the 4th level of images.
decomposition, which highlights pretty much the quality of
the fused image. In that we can see values that have reached IV. C ONCLUSION AND F UTURE S COPE
a higher value, especially the entropy.Since the results can be The wavelet based medical image fusion using various
changed depending on the quality of the image to be fused, we fusion rules is applied on the CT and MRI images of the

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.
 IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

TABLE I
P ERFORMANCE C OMPARISON OF WAVELET FUNCTIONS OF IMAGE FUSION

Level
of De-
Fusion
Wavelet com- MI SD Entropy
Rule
pos-
tion
Haar 5 (max,max) 0.714604 49.26077 5.822583
db1 5 (max,max) 0.714604 49.26077 5.822583
db2 5 (max,max) 0.704004 48.93627 5.95771
db3 5 (max,max) 0.702615 48.9583 5.997283
db4 5 (max,max) 0.7013 48.94462 6.04485
db5 5 (max,max) 0.700982 48.6933 6.050444
db6 5 (max,max) 0.70036 48.86898 6.061784
db7 5 (max,max) 0.698702 48.75792 6.081972
db8 5 (max,max) 0.696402 48.66749 6.105508
db9 5 (max,max) 0.695614 48.6405 6.099863
db10 5 (max,max) 0.697709 48.61308 6.124046
sym2 5 (max,max) 0.704004 48.93627 5.95771
sym3 5 (max,max) 0.702615 48.9583 5.997283
Fig. 9. Fused Image at Level of decomposition 2
sym4 5 (max,max) 0.700843 49.01287 6.00862
sym5 5 (max,max) 0.705984 48.97472 5.979811
sym6 5 (max,max) 0.700525 48.97258 6.019689
sym7 5 (max,max) 0.702949 48.96526 6.020996
sym8 5 (max,max) 0.699826 48.89905 6.04608
coif1 5 (max,max) 0.70528 49.0809 5.934525
coif2 5 (max,max) 0.700927 48.96451 6.013807
coif3 5 (max,max) 0.700625 49.00378 6.04093
coif4 5 (max,max) 0.699539 48.85999 6.055085
coif5 5 (max,max) 0.69928 48.73318 6.063631
bior1.1 5 (max,max) 0.714604 49.26077 5.822583
bior1.3 5 (max,max) 0.709188 49.49946 5.975967
bior1.5 5 (max,max) 0.706217 49.75647 6.008838
bior2.2 5 (max,max) 0.730418 50.82155 5.957674
bior2.4 5 (max,max) 0.7216 50.64709 6.042627
bior2.6 5 (max,max) 0.718495 50.73492 6.063203
bior2.8 5 (max,max) 0.720261 50.69373 6.051023
bior3.1 5 (max,max) 0.74614 60.28254 6.168947
bior3.3 5 (max,max) 0.739808 55.17932 6.166977
bior3.5 5 (max,max) 0.733493 54.02874 6.139239
bior3.7 5 (max,max) 0.733556 53.5092 6.118029
Fig. 10. Fused Image at Level of decomposition 3
bior 3.9 5 (max,max) 0.726682 53.11953 6.171972
bior4.4 5 (max,max) 0.702704 48.93388 5.984741
bior5.5 5 (max,max) 0.695298 48.50159 5.947795
human brain. The information in the soft tissue and bony
bior6.8 5 (max,max) 0.703143 49.13668 6.03986
structure both are important in diagnosis of any tumor or
rbio1.1 5 (max,max) 0.714604 49.26077 5.822583
rbio1.3 5 (max,max) 0.709712 48.99245 5.905848
any fracture. Now the fusion is done using different wavelet
transforms which are the finest methods of transformation
TABLE II
technique in image processing. From our experimental analysis
CHANGE IN PERFORMANCE OF FUSION AS RESULT OF LEVEL OF it is found out that biorthogonal (bior3.1) wavelet is the
DECOMPOSITION best one for fusion transformation which can be used for
Level of
the biomedical image fusion purposes. This transform with
Fusion 5 levels of decomposition (as the number of decompositions
Wavelet Decom- MI SD Entropy
Rule
postion increases the fusion becomes more effective) and max-max
bior3.1 1 (max,max) 0.780111 49.13172 5.714862 rule gives encouraging results in terms of all evaluation
bior3.1 2 (max,max) 0.777907 50.95935 5.865336 metrics. Also, among all the fusion rules, the ’maximum’
bior3.1 3 (max,max) 0.765978 53.16807 6.030338
fusion rule shows better results than all other rules like mean,
bior3.1 4 (max,max) 0.756239 56.08111 6.184075
minimum and random etc.
bior3.1 5 (max,max) 0.74614 60.28254 6.168947
As it is observed that bior3.1 is showing better results in
fusion, if it is optimized [19] further using any kind of meta-
heuristic algorithm [20] or dictionary learning approach, there

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.
 IEEE 7th International Conference on Smart Structures and Systems ICSSS 2020

[8] Y. Yang, D. S. Park, S. Huang, and N. Rao, “Medical image fusion via
an effective wavelet-based approach,” EURASIP Journal on Advances
in Signal Processing, vol. 2010, pp. 1–13, 2010.
[9] R. Singh, S. Nigam, A. K. Singh, and M. Elhoseny, “An overview of
medical image fusion in complex wavelet domain,” in Intelligent Wavelet
Based Techniques for Advanced Multimedia Applications. Springer,
2020, pp. 31–50.
[10] A. Petrosian and F. Meyer, Wavelets in Signal and Image Analysis: From
Theory to Practice, ser. Computational Imaging and Vision, A. Petrosian
and F. Meyer, Eds. Springer Netherlands, 2013.
[11] R. V. Ravi and K. Subramaniam, “Image compression and encryption
using optimized wavelet filter bank and chaotic algorithm,” International
Journal of Applied Engineering Research, vol. 12, no. 21, pp. 10 595–
10 610, 2017.
[12] H. Li, B. Manjunath, and S. K. Mitra, “Multisensor image fusion using
the wavelet transform,” Graphical models and image processing, vol. 57,
no. 3, pp. 235–245, 1995.
[13] P. Zeeuw, “Wavelet and image fusion, cwi,” Amsterdam, The Nether-
lands, March, 1998.
[14] S. M.V, S. K. M, T. A. Asharaf, S. C.T, S. M. M.T, and R. V.
Ravi, “Biomedical image fusion in wavelet domain; a brief survey,”
International Journal of Innovative Science, Engineering & Technology,
2015.
Fig. 11. Fused Image at Level of decomposition 4 [15] M. Hossny, S. Nahavandi, and D. Creighton, “Comments on ’informa-
tion measure for performance of image fusion’,” Electronics Letters,
vol. 44, no. 18, pp. 1066–1067, 2008.
[16] Guihong Qu, Dali Zhang, and Pingfan Yan, “Information measure for
performance of image fusion,” Electronics Letters, vol. 38, no. 7, pp.
313–315, 2002.
[17] V. Naidu and J. R. Raol, “Pixel-level image fusion using wavelets and
principal component analysis,” Defence Science Journal, vol. 58, no. 3,
p. 338, 2008.
[18] R. Singh and A. Khare, “Fusion of multimodal medical images using
daubechies complex wavelet transform–a multiresolution approach,”
Information Fusion, vol. 19, pp. 49–60, 2014.
[19] R. V. Ravi and K. Subramaniam, “Optimized wavelet filters and modified
huffman encoding-based compression and chaotic encryption for image
data,” International Journal of Applied Engineering Research, vol. 12,
no. 13, pp. 3961–3977, 2017.
[20] R. V. Ravi, K. Subramaniam, T. Roshini, S. P. B. Muthusamy, and G. P.
Venkatesan, “Optimization algorithms, an effective tool for the design of
digital filters; a review,” Journal of Ambient Intelligence and Humanized
Computing, pp. 1–17, 2019.

Fig. 12. Fused Image at Level of decomposition 5

may be a chance for improving the results.

R EFERENCES
[1] J. Agarwal and S. S. Bedi, “Implementation of hybrid image fusion
technique for feature enhancement in medical diagnosis,” Human-centric
Computing and Information Sciences, vol. 5, no. 1, pp. 1–17, 2015.
[2] H. M. El-Hoseny, W. Abd Elrahman, E. Rabaie, M. El-Sayed, O. S.
Faragallah, A. El-Sami, and E. Fathi, “Medical image fusion: A literature
review present solutions and future directions,” Menoufia Journal of
Electronic Engineering Research, vol. 26, no. 2, pp. 321–350, 2017.
[3] S. Bedi, J. Agarwal, and P. Agarwal, “Image fusion techniques and
quality assessment parameters for clinical diagnosis: a review,” Inter-
national journal of advanced research in computer and communication
engineering, vol. 2, no. 2, pp. 2319–5940, 2013.
[4] S. Li, X. Kang, L. Fang, J. Hu, and H. Yin, “Pixel-level image fusion: A
survey of the state of the art,” information Fusion, vol. 33, pp. 100–112,
2017.
[5] H. Mitchell, Image Fusion: Theories, Techniques and Applications,
H. Mitchell, Ed. Springer Berlin Heidelberg, 2010.
[6] S. Jana, A. Pal, S. Ray, and P. Adhikary, “A study on wavelet based
image fusion,” pp. 1–4, 2018.
[7] A. V. Vanmali, T. Kataria, S. G. Kelkar, and V. M. Gadre, “Ringing
artifacts in wavelet based image fusion: Analysis, measurement and
remedies,” Information Fusion, vol. 56, pp. 39–69, 2020.

978-1-7281-7223-1/20/$31.00©2020 IEEE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on September 29,2024 at 17:47:08 UTC from IEEE Xplore. Restrictions apply.