International Journal of Research in Advent Technology, Vol.2, No.

3, March 2014
E-ISSN: 2321-9637

Multimodal Image Fusion Techniques for Medical

Images using Wavelets
Deron Rodrigues1, Hasan Ali Virani2, Shajahan Kutty3
Electronics and Telecommunication Department, Goa College of Engineering, Goa1, 2, 3
Student, ME– ETC- ECI 1, Associate Professor2, Assistant Professor3
Email: [email protected]

Abstract- Image fusion has become crucial part of medical diagnosis. This paper gives introduction to image
fusion methods based on wavelet transform. Fusion of CT scanned images and MRI images using multi
resolution wavelet transform with necessary preprocessing of it is proposed. It also compares the performance of
the various types of wavelet basis families used and the different fusion rules used to fuse the approximation and
detail wavelet coefficients.

Index Terms- Image Fusion, Discrete Wavelet Transforms, CT, MRI, multimodal medical Image fusion,
Medical imaging, Fusion performance metrics, entropy, Mean Square Error, Peak Signal to Noise Ratio

1. INTRODUCTION: advantages and disadvantages. Major advantage of the

The process of combining the relevant information CT is its ability to image bone and blood vessels all at
from a set of images into a single superior quality the same time. In the CT scanned image of the brain
fused image is called as image fusion. This high hard tissue like the skull bone is clearly seen but the
quality fused image needs to have better spatial and soft tissue like the membranes covering the brain are
spectral information, than any of the input images less visible and hence CT is not sensitive in detecting
[4][2]. inflammation of the membranes in the brain. In the
Image fusion method can be broadly classified into MRI scanned image of the same brain we observe the
two groups [5] – soft tissue like the membranes covering the brain can
1. Spatial domain fusion method be clearly seen but the hard tissue like the skull bones
2. Transform domain fusion. cannot be clearly seen. [10] These are the demerits of
In spatial domain techniques, the pixel values are the CT and MRI scans. Combination of CT and MRI
manipulated to achieve desired result. The Spatial scanned images of the brain then yields an image in
domain fusion methods include averaging, Brovey which both hard tissue like skull bones and the soft
method, principal component analysis (PCA), tissue like the membranes covering the brain can be
Intensity Hue saturation (HIS) based methods and clearly visible. The main purpose of medical image
high pass filtering based technique. The spatial fusion is to obtain a high resolution image which
domain approaches is that tend to produce spatial contains as much details as possible for the sake of
distortion and Spectral distortion in the fused image, diagnosis. So if these two images of the same organ
hence prove a disadvantage. In transform domain are fused then the fused image contains considerably
methods the image is first transferred into another more amount of information for diagnosis of that
domain, say frequency domain and all the fusion organ, as compared to the non-fused images.
operations are performed on the transform of the
image and then the inverse transform is performed to 2. DISCRETE WAVELET TRANSFORMS
get the resultant image. Spatial distortion can be very Wavelets are finite duration oscillatory functions with
well handled by frequency domain approaches on zero average value. They have finite energy and hence
image fusion. However, its excellent characteristic in are suited for analysis of transient signal. [4] Wavelets
one-dimension can’t be extended to two dimensions can be described by using two functions namely the
or multi-dimension simply. The multi resolution scaling function f(t) or father wavelet and the wavelet
analysis has become a very useful tool for analyzing function ψ(t) or mother wavelet.[7][8]
remote sensing images, medical images etc. The A number of basis functions can be used as the
discrete wavelet transform is now becoming a very mother wavelet for Wavelet Transformation. The
important tool for fusion. mother wavelet through translation and scaling
Image Fusion techniques play important role in produces various wavelet families which are used in
medical imaging, microscopic imaging, remote the transformation. The wavelet families are given by
sensing, computer vision and robotics. In medical following equation numbered 1,
field, the different modalities like X-rays, CT scan,
MRI Scan, nuclear medicine and other modalities are …… (1)
used to examine the organs of body have their own

310
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

The Haar, Daubechies, Symlets and Coiflets are hence are needed to be registered before they are
compactly supported orthogonal wavelets. These fused.
wavelets along with Meyer wavelets are capable of
perfect reconstruction. The Meyer, Morlet and 3.1.2. Image Resizing
Mexican Hat wavelets are symmetric in shape. The
wavelets are chosen based on their shape and their Also the sizes of the images might vary so before
ability to analyze the signal in a particular application. fusion, the images are needed to be resized so that
The Discrete Wavelet Transform has the property both the images are of the same size. This is done by
that the spatial resolution is small in low-frequency interpolating the smaller size image by rows and
bands but large in high-frequency bands. This is columns duplication.
because the scaling function is treated as a low pass
filter and the mother wavelet as high pass filter in 3.1.3 Image Enhancement
DWT implementation. [4]
The wavelet transform decomposes the image into If both or any of the images are not of grayscale then
low-high, high-low, high-high spatial frequency bands it is desired that it is converted to grayscale. The next
at different scales and the low-low band at the step which follows this is equalization of the
coarsest scale which is shown in figure 1(a). The low- histograms of the images so that the contrast of the
low image has the smallest spatial resolution and image is enhanced and that both the images have
represents the approximation information of the similar range of values for wavelet coefficients.
original image. The other sub-images, on the contrary
show the detailed information of the original image. 3.2 Image Fusion Algorithm
There are different levels of decomposition which The steps in the algorithm for image fusion using
are shown in Figure 1. After one level of DWT as shown in Figure 2 are as follows:
decomposition, there will be four frequency bands, as (1) Read the input images (MRI & CT Scanned).
listed above. The next level decomposition is just (2) Resample and register both these images.
applied to the LL band of the current decomposition (3) Apply 2D-discrete wavelet transform to these
stage, which forms a recursive decomposition images which decompose it into four sub-bands
procedure. Thus, N-level decomposition will finally (LL, LH, HL and HH).
have 3N+1 different frequency bands, which include (4) The Wavelet coefficients obtained from both the
3N high frequency bands and just one LL frequency images are fused using the rules for fusion.
band. [12] (5) The final fused image is reconstructed by applying
inverse discrete wavelet transform to fused image.

Fig. 1. Wavelet Transform (a) Single Level decomposition (b) Two

level decomposition (c) three level decomposition

3. IMAGE FUSION USING DWT

3.1 Preprocessing for Image Fusion

The multimodal images which are needed to be fused
need to be processed prior to application of fusion
algorithm. [8] The pre-processing includes image
registration, image resizing and histogram
equalization.

3.1.1. Image Registration

The images which are obtained by different
modalities might be of different orientations and
Fig. 2. Block diagram of Image Fusion process using DWT

311
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

3.3 Fusion Rules 5. RESULTS:

By application of Wavelet Transforms to the images The image fusion was carried over for multimodal
we obtain the Wavelet Coefficients which are called images obtained from CT and MRI scans. The
as approximation coefficients, horizontal detail wavelet used was second order symlet and different
coefficients, vertical detail & diagonal detail types of fusion rules were used for the same set of
coefficients. [6]These corresponding coefficients of images. After obtaining the synthesized image the
each of the image are to be fused together in a performance evaluation of the images was carried out.
particular manner.
Certain rules that can be used for the same are
Maximum value, Minimum value, Mean Value and
Random value. These techniques respectively merge
the two approximations or details structures obtained
from the images to be fused, element wise by taking
the maximum, minimum, mean or a randomly chosen
element. Besides these there are other rules of fusing
the wavelet coefficients such as linear fusion rule, Up- Fig. 4.a CT scan image Fig. 4.b MRI Scan Image
down fusion rule, down-up fusion rule, left-right
fusion rule and right-left fusion rule.

4. PERFORMANCE METRICS TO EVALUATE

IMAGE FUSION TECHNIQUES:

The image fusion process should preserve all valid

and useful information from the input images, so also Fig. 4.c Max-Max Fused image Fig. 4.d Min-Min Fused Image
it should not introduce undesired artifacts. Various
metrics are used in order to evaluate the performance
of Image Fusion techniques. [4] They are as follows:

4.1. Entropy:
Entropy measures the information quantity contained
in an image. [7][11] Higher entropy value of the fused Fig. 4.e. Mean-Mean Fused image Fig. 4.f. Lin-Lin Fused image
image indicates presence of more information and
improvement in fused image. If L is the total of grey
levels and p= {p0, p1,…..,pL-1} is the probability
distribution of each level, Entropy is defined as,

Fig. 4. g Min-Max Fused image Fig.4.h. Lin-Max Fused image

4.2. Mean Square Error:
The MSE between image X, and an approximation, Y,
is the squared norm of the difference divided by the
number of elements in the image. If i and j are pixel
row column indices, M and N are the number of rows
and columns, MSE is defined by,
Fig. 4.i. Down_Up-Down_Up Fig. 4.j. Down_Up-Up_Down
Fused image Fused image

4.3 .Peak Signal to Noise Ratio:

PSNR is the ratio between the maximum possible
power of a signal and the power of corrupting noise
that affects the fidelity of the signal. It is defined by,

Fig. 4.k Up_Down Down_Up Fig. 4.l Up_Down- Up_Down

Fused image Fused image

312
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

The synthesized images are shown in Figure 4. The Engineering Research March. 2013, Vol. 2 Iss. 1,
figures 4.a, b show the input CT scanned image and PP. 30-36
MRI scanned image respectively. [3] Chetan K. Solanki, Narendra M. Patel, Pixel
Fig. 4.c,d,e,f show synthesized image having fused based and Wavelet based Image fusion Methods
the both approximation and detail coefficients using with their Comparative Study, National
maximum fusion rule, minimum fusion rule, mean Conference on Recent Trends in Engineering &
fusion rule, linear fusion rule respectively. Whereas Technology, May 2011
Figures 4.g,h, show image fused by using minimum [4] Deepak Kumar Sahu, M.P. Parsai, Different
and linear fusion rule for approximation coefficients Image Fusion Techniques-Critical Review,
respectively and maximum fusion rule for detail International Journal of Modern Engineering
coefficients. Figure 4. i,j,k,l show combinations of Research (IJMER) www.ijmer.com Vol. 2, Issue.
up_down and down_up fusion rules. 5, Sep.-Oct. 2012 pp-4298-4301
The results obtained for entropy are displayed in the [5] Firooz Sadjadi, Comparative Image Fusion
table 1. Analysis,
Table 1 Image Fusion rules and corresponding Entropy.
[6] Hui Li, B.S. Manjunath and Sanjit K. Mitra,
Multi-sensor Image fusion using the Wavelet
Fusion Rule Entropy Transform.
(bits) [7] Kokare Swati, Premanand Kadbe, Biomedical
Original CT Scanned image 6.3070 Image Fusion Using Wavelet And Curvelet
Original MRI Scanned image 5.6294 Transform, IJERT,Vol. 2 Issue 5, May – 2013
Max-Max Fused image 7.1072 [8] R.J.Sapkal, S.M.Kulkarni, Image Fusion based on
Min-Min Fused Image 6.4442 Wavelet Transform for Medical Application,
Mean-Mean Fused image 6.9795 IJERA, Vol. 2, Issue 5, September- October
Linear-Linear Fused image 6.9849 2012, pp.624-627
Min-Max Fused image 6.6570 [9] Shih-Gu Huang, “Wavelet for Image Fusion”
Lin-Max Fused image 7.0746 [10] Suresh Padala Fusion of CT and MRI Scanned
Down_Up-Down_Up Fused image 7.0227 Medical Images Using Image Processing.
Down_Up-Up_Down Fused image 7.0378 [11] Wenzhong Shi, Chang Qing Zhu, Yan Tian, Janet
Up_Down Down_Up Fused image 7.0378 Nichol, Wavelet-based image fusion and quality
Up_Down- Up_Down Fused 7.1133 assessment, International Journal of Applied
image Earth Observation and Geoinformation 6 (2005)
241–251
[12] Yong Yang, Dong Sun Park, and Shuying Huang,
6. CONCLUSION:
Medical Image Fusion via an Effective Wavelet
The synthesized image has the merits of both CT & based Approach.
MRI scanned images. In this paper, the comparison of
various fusion rules to fuse wavelet coefficients in the
image fusion process is carried out. In this paper, 10
kinds of wavelet-based fusion methods are compared.
The Entropy (EN) has been used as image fusion
evaluation criterion for quantitative analysis of image
fusion results. The Matlab fusion results indicate that
the fusion methods produced higher entropy, the
highest being produced by Up_Down- Up_Down
technique. The comparative analysis of image fusion
techniques allows in selecting the best fusion method
and therefore one can obtain better visualization of the
fused image.

REFERENCES

[1] Anjali A. Pure, Neelesh Gupta, Meha

Shrivastava, An Overview of Different Image
Fusion Methods for Medical Applications ,
IJSER, Volume 4, Issue 7, July-2013
[2] Changtao He, Guiqun Cao, Fangnian Lang, An
Efficient Fusion Approach for Multispectral and
Panchromatic Medical Imaging, Biomedical

313