ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 9(22), DOI: 10.17485/ijst/2016/v9i22/95298, June 2016 ISSN (Online) : 0974-5645

Comparative Analysis of Similarity Measure

Performance for Multimodality Image Fusion using
DTCWT and SOFM with Various Medical Image
Fusion Techniques
C. Karthikeyan1* and B. Ramadoss2
Department of CSE, Jawaharlal Nehru Technological University, Hyderabad, India;
1

[email protected]
2
Department of Computer Applications, National Institute of Technology, Tiruchirappalli, India;
[email protected]

Abstract
Objectives: In this paper, the performance of similarity measures such as Edge Based Similarity Measure and Structural
Similarity Index Measure is evaluated and also compared with the existing medical image fusion techniques. Materials
and Methods: Multimodality Medical Image fusion is the process of fusing two Medical images obtained from two different
sensors for better diagnosis. Medical image fusion combines and merges all relevant and complementary information from
multiple source images into single composite image which facilitates more precise diagnosis and better treatment. The fused
image should convey a better description of the scene than the individual images. The performance of the fused image is
evaluated by various metrics such as Peak Signal to Noise Ratio (PSNR), Entropy, Standard deviation, Edge Based Similarity
Measure (EBSM) and Structural Similarity Index Measure (SSIM). This paper proposes a method for fusion of Medical images
using Dual Tree Complex Wavelet Transform (DTCWT) and Self Organizing Feature Map (SOFM). Findings: The performance
of the proposed fusion algorithm is evaluated over pairs of CT and MR images obtained from patients in comparison with
existing fusion techniques such as Discrete Wavelet Transform (DWT), Nonsubsampled Contourlet Transform (NSCT) and
Fast Discrete Curvelet Transform (FDCT). In this paper, performance is evaluated by using the metric; Edge based Similarity
Measure (EBSM), and Structural Similarity Index (SSIM). Applications / Improvements: Through the simulation result,
as compared with the DWT, FDCT, NSCT and DTCWT fusion methods, it is concluded that the Multimodality image fusion
using DTCWT with Robust Second Order First Moment (SOFM) gives better Edge based similarity measure and Structural
similarity index measure.

Keywords: Dual Tree Complex Wavelet Transform, Edge Based Similarity Measure, Fuzzy Rules, Structural Similarity Index

An efficient image fusion method using wavelet combined

1. Introduction
transformation is implemented in2 for multi sensor lunar
Image fusion combines two or more than two images into image data. A new information theoretic fusion algorithm
a single composite image. Obviously the composite image combined neural network and fuzzy theory is presented
should give a better composition of the scene than the in3-4. Image fusion method based on non subsampled con-
individual ones. A novel approach of multimodal medi- tour let transform can achieve better fusion performance5.
cal image fusion using wavelet transforms is presented in1. This method is complex in fusion algorithm and also lack

*Author for correspondence

 Comparative Analysis of Similarity Measure Performance for Multimodality Image Fusion using DTCWT and SOFM with Various
Medical Image Fusion Techniques

of multi resolution feature. An innovative multilevel image

fusion algorithm using Fast Discrete Curvelet Transform
gives the best fusion result in terms of enhanced visual
quality in fused images6. But it suffers from multi direc-
tional decomposition.
The DWT is the good method for image fusion. DWT
and spatial frequency based image fusion algorithm is
discussed in7. A DWT based multimodal medical image
fusion is presented in8. DWT has two disadvantages, lack
of shift invariance and poor directional selectivity. Dual
Tree Complex Wavelet transform overcome the disad-
vantages of DWT. DTCWT based multimodal medical
image fusion is implemented in9. In this paper, an efficient
approach for medical image fusion based on DTCWT
and SOFM is proposed. The importance of Multimodal
Medical image fusion is discussed in section 2. The pro-
posed image fusion algorithm is explained in section 3.
The experimental results obtained from the proposed
fusion technique are discussed in section 4. In section 5, Figure 1. Blockdiagram of the Proposed Image Fusion.
conclusion is made from the results obtained.

2. Medical Image Fusion sensors like Computed Tomography (CT) and Magnetic
Resonance Imaging (MRI) has been considered in this
Medical imaging has become an important factor in work. Many techniques have been proposed in the litera-
diagnosis, treatment and research. Medical image fusion ture has been studied and reviewed. Proposed method is
combines and merges all relevant and complementary demonstrated in Figure 1.
information from multiple source images into single Image Registration is the process of establishing spa-
composite image. In Medical image fusion, the fusion of tial correspondence between two or more images of the
images can often lead to additional clinical information same scene taken at different times or from different view-
not apparent in the separate images. The requirements points or by different sensors. In this work, the intensity
of Medical image fusion are that the fused image should based image registration done in MRI to position same
convey more information than the individual images and coordination.
should not introduce any artifacts or distortions. After image registration, the registered image and the
A review about various image fusion algorithms image to be fused are decomposed by using DTCWT at
based on Discrete Wavelet Transform (DWT), Principal predefined scale14. The robust SOFM neural network12
Component Analysis (PCA), morphological methods, is utilized to recognize and extract the features. This can
knowledge based methods, neural network based meth- be done by training and simulating the network for the
ods and fuzzy logic method and their performances are resultant coefficients (approximation and detailed) of
compared in10. Fusion rule also plays a vital role in the each level of MR and CT images. After decomposition,
image fusion. The rules like min rule, max rule, fuzzy the proposed method is carried out on each sub band
rule11 can be applied and are seen in the literature. independently.
In order to identify the salient region in each sub-
3. Multimodality Image Fusion band, clustering13 based image thresholding is applied. It
using DTCWT and SOFM assumes that the input image has bi-model histogram and
calculates optimum threshold value based on their intra
The proposed multimodality image fusion technique class variance14. Let us consider A, B are the registered
is based on the DTCWT and SOFM. The fusion of image and the image to be fused respectively. AA and BB
Multimodal Medical images obtained from different are their corresponding threshold image. Then, the fusion

2 Vol 9 (22) | June 2016 | www.indjst.org Indian Journal of Science and Technology
 C. Karthikeyan and B. Ramadoss

process after thresholding the images for all wavelet Table 1. EBSM Performance of the proposed method
coefficients located at pixel position (i,j) : Image Edge based similarity measure
Set DWT FDCT NSCT DTCWT DTCWT-
Input : Registered image (A) and corresponding clus- SOFM
tered image (AA), image to be fused (B) and
1 0.2476 0.3643 0.4513 0.5677 0.8136
corresponding clustered image (BB)
2 0.2564 0.407 0.5078 0.6151 0.8545
Output : Fused Image (F)
Rule1 : if AAi,j is true and Bbi,j is false then Fi,j = Ai,j 3 0.2599 0.3704 0.4703 0.5925 0.8241
Rule2 : if Aai,j is false and Bbi,j is true then Fi,j = Bi,j 4 0.2866 0.3653 0.4555 0.5589 0.8065
Rule3 : if Aai,j is true and Bbi,j is true then Fi,j = 5 0.3271 0.3696 0.4405 0.5265 0.7823
max(Ai,j, Bi,j)
Rule4 : if Aai,j is false and Bbi,j is false then Fi,j =
min(Ai,j, Bi,j)

Then apply Inverse DTCWT to the fusion result to get

the final fused image.

4. Results and Discussions

The performance of the proposed method is evaluated
over pairs of CT and MR images obtained from patients
in comparison with existing fusion techniques such as
DWT, NSCT, FDCT. In this paper, performance is evalu-
ated by using the metric; Edge Based Similarity Measure
(EBSM), and Structural Similarity Index (SIM).
Figure 2. Visual results of the proposed DTCWT-SOFM
4.1 Edge Based Similarity Measure (EBSM) for five pairs of MRI and CT images in comparison with
existing fusion techniques.
The edge based similarity measure gives the similarity
between the edges transferred in the fusion process. methods based on DWT, FDCT, NSCT. The reason for
Mathematically, is defined Q AB / F is defined as better performance of the proposed fusion method is that
M N the proposed fusion is the mixture of min rule and max
∑∑ Q
=i 1 =j 1
AF
i, j wix, j + QiBF y

, j wi , j  rule and the application of these rules are based on the
Q AB / F = M N clustering based thresholding approach. Table 1 shows
∑∑ w x
i, j + wiy, j  the Comparison on Edge Based Similarity Measure of dif-
=i 1 =j 1
(1) ferent methods for Fused Image. Figure 2 shows the visual
Where A, B and F represent the input and fused results of the proposed fusion method for five pairs of
images respectively. images in comparison with existing fusion methods and
The definition of Q AF and Q BF are same and given as Figure 3 shows the Chart representation of Comparison
on Edge Based Similarity Measure of different methods
QiAF AF AF
, j = Q g ,i , j Qα ,i , j , for Fused Image.
QiBF BF BF
, j = Q g ,i , j Qα ,i , j (2)
*F 4.2 Structural Similarity Index Measure
where Q g*F and Qα are the edge strength for images A
and B.
(SSIM)
It is observed from the Table 1 that the performance The structural similarity index is a method for measuring
of the proposed method outperforms the existing the similarity between two images using eqn. (3).

Vol 9 (22) | June 2016 | www.indjst.org Indian Journal of Science and Technology 3
 Comparative Analysis of Similarity Measure Performance for Multimodality Image Fusion using DTCWT and SOFM with Various
Medical Image Fusion Techniques

Figure 3. Comparison on Edge Based Similarity Measure Figure 4. Comparison on Structural Similarity Index
(EBSM) of different techniques for Fused Image. Measure (SSIM) of different techniques for Fused Image.

Table 3. Percentage of improvement on EBSM and

Table 2. Performance evaluation of SSIM using DTCWT- SSIM for DWT vs. proposed method
SOFM and Fuzzy rule based fusion method
Proposed
Structural Similarity Index Measure ( SSIM ) Image % of
Metrics DWT DTCWT-
Set improvement
Image DTCWT- SOFM
Set DWT FDCT NSCT DTCWT SOFM 1 0.2476 0.8136 228.59
1 0.321 0.343 0.451 0.577 0.831
Edge based 2 0.2564 0.8545 233.27
2 0.364 0.407 0.467 0.566 0.857 similarity 3 0.2599 0.8241 217.08
3 0.359 0.377 0.472 0.555 0.835 measure 4 0.2866 0.8065 181.40
4 0.327 0.359 0.467 0.567 0.839 5 0.3271 0.7823 139.16
5 0.336 0.375 0.445 0.555 0.783 1 0.321 0.831 158.88
Structural 2 0.364 0.857 135.44
Similarity
3 0.359 0.835 132.59
Index
(2µ x µ y + c1 )(2σ xy + c2 ) Measure 4 0.327 0.839 156.57
SSIM (x , y ) = 5 0.336 0.783 133.04
(µ 2x + µ 2y + c1 )(σ2x + σ2y + c2 ) (3)
with
mx the average of x; my the average of y; σ2x the vari-
2
ance of x; σ y the variance of y; 4.3 Comparative Analysis
σxy the covariance of x and y ; The proposed image fusion method using DTCWT with
Table 2 shows the Comparison on Structural Similarity SOFM is quantitatively and qualitatively compared in
Index Measure of different methods for Fused Image. terms of EBSM, SSIM and image quality with four other
It is observed from the Table 2 that the performance of existing state of the art methods for image fusion. Table 3
the proposed fusion algorithm outperforms the existing – Table 6 shows the percentage of improvement for EBSM
algorithms based on DWT, FDCT, NSCT and DTCWT. and SSIM between proposed multimodality image fusion
The reason for better performance of the proposed fusion using DTCWT with SOFM and existing state of the art
method is that the proposed fusion is the mixture of min methods; DWT, FDCT, NSCT and DTCWT.
rule and max rule and the application of these rules are Table 3 gives the percentage of improvement on
based on the clustering based thresholding approach. EBSM and SSIM between proposed multimodality image
Table 2 shows the Comparison on Structural Similarity fusion and DWT. From Table 3, the simulation results
Index Measure of different methods for Fused Image. show that the proposed DTCWT and SOFM method
Figure 4 shows the Chart representation of Comparison has achieved maximum edge based similarity measure of
on Structural Similarity Index Measure of different methods 0.8545 and structural similarity index measure of 0.831
for Fused Image. whereas the existing DWT method with maximum Edge

4 Vol 9 (22) | June 2016 | www.indjst.org Indian Journal of Science and Technology
 C. Karthikeyan and B. Ramadoss

Table 4. Percentage of improvement on EBSM and SSIM Table 5. Percentage of improvement on EBSM and
for FDCT vs. Proposed Method SSIM for NSCT vs. Proposed Method
Proposed Proposed
Image % of Image % of
Metrics FDCT DTCWT- Metrics NSCT DTCWT-
Set improvement Set improvement
SOFM SOFM
1 0.3643 0.8136 123.33 1 0.4513 0.8136 80.28
2 0.407 0.8545 109.95 Edge based 2 0.5078 0.8545 68.27
Edge based
similarity 3 0.3704 0.8241 122.49 similarity 3 0.4703 0.8241 75.23
measure measure 4 0.4555 0.8065 77.06
4 0.3653 0.8065 120.78
5 0.3696 0.7823 111.66 5 0.4405 0.7823 77.59
1 0.343 0.831 142.27 1 0.451 0.831 84.26
Structural Structural 2 0.467 0.857 83.51
2 0.407 0.857 110.57
Similarity Similarity
3 0.377 0.835 121.49 3 0.472 0.835 76.91
Index Index
4 0.359 0.839 133.70 Measure 4 0.467 0.839 79.66
Measure
5 0.375 0.783 108.80 5 0.445 0.783 75.96

Table 6. Percentage of improvement on EBSM and

SSIM for DTCWT vs. Proposed Method
based similarity measure of 0.2564 and structural similarity
index measure of 0.321. Thus there is an improvement of Proposed
Image % of
233.27% in edge based similarity measure and 158.88% in Metrics DTCWT DTCWT-
Set improvement
structural similarity index measure. SOFM
Table 4 gives the percentage of improvement on EBSM 1 0.5677 0.8136 43.32
and SSIM between proposed multimodality image fusion Edge based 2 0.6151 0.8545 38.92
method and FDCT. similarity 3 0.5925 0.8241 39.09
From Table 4, the simulation results show that the measure 4 0.5589 0.8065 44.30
proposed DTCWT and SOFM method has achieved 5 0.5265 0.7823 48.58
maximum edge based similarity measure of 0.8136 and 1 0.577 0.831 44.02
structural similarity index measure of 0.831 whereas the Structural 2 0.566 0.857 51.41
existing FDCT method with maximum Edge based simi- Similarity
3 0.555 0.835 50.45
larity measure of 0.3643 and structural similarity index Index
Measure 4 0.567 0.839 47.97
measure of 0.343. Thus there is an improvement of 123.33
in edge based similarity measure and 142.27% in struc- 5 0.555 0.783 41.08
tural similarity index measure.
Table 5 gives the percentage of improvement on EBSM results show that the proposed DTCWT-SOFM method
and SSIM between proposed technique and NSCT. From has achieved maximum edge based similarity measure of
Table 5, the simulation results show that the proposed 0.7823 and structural similarity index measure of 0.857
DTCWT and SOFM method has achieved maximum whereas the existing DTCWT method with maximum
edge based similarity measure of 0.8136 and structural Edge based similarity measure of 0.5265 and struc-
similarity index measure of 0.831 whereas the existing tural similarity index measure of 0.566. Thus there is an
NSCT method with maximum Edge based similarity improvement of 48.58% in edge based similarity measure
measure of 0.4513 and structural similarity index mea- and 51.41% in structural similarity index measure.
sure of 0.451. Thus there is an improvement of 80.28 in
edge based similarity measure and 84.26% in structural
5. Conclusion
similarity index measure.
Table 6 gives the percentage of improvement on EBSM In Medical image processing applications, specifically
and SSIM between proposed multimodality image fusion in MRI and CT images, the edge preserve is an impor-
method and DTCWT. From Table 6, the simulation tant criterion in complementary details of input images.

Vol 9 (22) | June 2016 | www.indjst.org Indian Journal of Science and Technology 5
 Comparative Analysis of Similarity Measure Performance for Multimodality Image Fusion using DTCWT and SOFM with Various
Medical Image Fusion Techniques

This work investigates the proposed Neural Network 5. Xing Su-xia, Lian Xiao-feng, Chen Tian-hua, Xiao
based nonlinear medical image fusion algorithm based Hongbing. Image Fusion Method Based on NSCT and
on DTCWT with Robust SOFM. This allows us to fuse Robustness Analysis. IEEE International Conference
two modalities, CT and MR images to visually assess the on Computer Distributed Control and Intelligent
details on a single image. The steps involved in the pro- Environmental Monitoring. 19-20 Feb 2011, p. 346−49.
6. Roshna J. Sapkal, Sunita M. Kulkarni, Innovative Multileve
posed fusion algorithm are image registration, DTCWT
Image Fusion Algorithm using combination of Transform
decomposition, Robust SOFM, image fusion and recon-
domain and spatial domain methods with comparative anal-
struction to obtain the fused image. The Similarity ysis of wavelet and curvelet transform. International Journal
Measure performance is evaluated by using Edge Based of Computer Applications. March 2013; 66(21):41-37.
Similarity Measure and Structural Similarity Index 7. Zhongxian QU, Jingbo Zhang, Zijun Feng. Image Fusion
Measure. The simulation results show that the proposed Algorithm Based on Two-Dimensional Discrete Wavelet
method has achieved maximum Edge based similarity Transform and Spatial Frequency. IEEE Fifth International
measure and Structural Similarity Index Measure values. Conference on Frontier of Computer Science and
Hence the fused image provides better diagnosis with- Technology. 18-22 Aug. 2010, p. 537−40.
out artifacts. Through the simulation result, as compared 8. Singh R, Khare A. A Wavelet Based Multimodal Medical
with the DWT, FDCT, NSCT and DTCWT fusion meth- Image Fusion, in Proceedings of International Symposium
ods, it is concluded that the Multimodality image fusion on Medical Imaging: Perspectives on Perception and
Diagnostics. 2010.
using DTCWT with Robust SOFM gives better Edge
9. Rajiv Singh, Richa Srivastava, Om Praksh, Ashish Khare.
based similarity measure and Structural similarity index
DTCWT Based Multimodal Medical Image Fusion.
measure.
Electronic Proceedings of the 13th British Machine Vision
Conference University of Cardiff, January 2012, p. 1−4.
6. References 10. Sahu, DK, Parsai MP. Different Image Fusion Techniques–A
Critical Review,” International Journal of Modern
1. Soma Sekhar, Giri Prasad MN. A Novel Approach of Image Engineering Research (IJMER). 2012; 2(5): 4298−4301.
Fusion on MR and CT Images Using Wavelet Transforms. 11. Heydar Jafarzadeh, Rouhollah Rahmati Torkashvand,
IEEE 3rd International Conference on Electronics Chamran Asgari, Amir Amiry. Provide a New Approach for
Computer Technology, 8-10 April 2011, p. 172−76. Mining Fuzzy Association Rules using Apriori Algorithm.
2. Lavanya A, Vani K, Sanjeevi S, Suresh Kumar R. Image Indian Journal of Science and Technology. 2015 Apr;
Fusion of the Multi sensor Lunar Image data using Wavelet 8(supp 7), p. 127−34. Doi:10.17485/ijst/2015/v8i8/63627
Combined Transformation. IEEE International Conference 12. Teuvo Kohonen. The self-organizing map. IEEE Proceedings.
on Recent Trends in Information Technology, 3-5 June 1990; 78(9):1464−80.
2011, p. 920−25. 13. Revathy S, Parvaathavarthini B, Rajathi S. Futuristic
3. De-gan Zhang, Yuan-yuan Wang, Wei-wei Liu, Xiao-li Validation Method for Rough Fuzzy Clustering. Indian
Zhang. A New Method for Image Fusion Based on Fuzzy Journal of Science and Technology. 2015 Jan; 8(2):120−27.
Neural Network. IEEE International conference on Doi: 10.17485/ijst/2015/v8i2/58943
Modelling Identification, 17-19 July 2010, p. 574−78. 14. Karthikeyan C, Ramadoss B. Fusion of Medical Images
4. Das S, Kundu M. A Neuro-Fuzzy Approach for Medical using Mutual Information and Intensity based Image
Image Fusion. IEEE Transactions on Biomedical Engineering. Registration Schemes. ARPN Journal of Engineering and
2013; 60(12):3347−53. Applied Sciences. May 2015; 10(8):3561−65.

6 Vol 9 (22) | June 2016 | www.indjst.org Indian Journal of Science and Technology