International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 129

ISSN 2229-5518

An Overview of Different Image Fusion Methods

for Medical Applications
Anjali A. Pure, Neelesh Gupta, Meha Shrivastava

Abstract— Image fusion is an important research topic in many related areas such as computer vision, remote sensing, robotics, and
medical imaging etc. Image fusion is the process of combining relevant information from several images into single image. The final output
image can provide more information than any of the single image. Now-a-days, almost all areas of medical diagnosis are impacted by the
digital image processing. For medical diagnosis, MRI and CT images are of main concern, both images give special sophisticated
characteristics of the organ to be imaged. Computed Tomography (CT) provides the best information on denser tissue like bones. Magnetic
Resonance Image (MRI) provides better information on soft tissues. With more available multimodality medical images in clinical
applications, the idea of combining images from different modalities become very important and medical image fusion has emerged as a
new promising research field that help physicians in the diagnosis process. More research has been done for fusion of MRI and CT images
using traditional wavelet transform and few attempts using curvelet transform. This paper provides an overview of different image fusion
methods for medical applications in brief and to improve the fusion results proposed method is described shortly.

Index Terms— CT image, Curvelet Transform, Fast discrete curvelet transform (FDCT), Image fusion, MRI image, Wavelet Transfor

——————————  ——————————

1 INTRODUCTION

IJSER
Image fusion is a tool to combine multimodal images by us-
ing image processing techniques. Specifically it aims at the Literature Survey reveals that, Shih-Gu Huang [1] proposed
integration of disparate and complementary data in order to different image fusion methods that have been developed to
enhance the information apparent in the images, as well as to perform image fusion. Wavelet transform used for image fu-
increase the reliability of the interpretation [1],[2]. Due to the sion produced superior results than different methods. S.
advent of new diseases complementary information are re- Vasuki, S. Gandhimathi et al. [2] proposed integration of
quired from different modalities. When sensitive organs like wavelets and PCA for fusion of medical images. Comparative
brain are scanned, both magnetic resonance imaging and analysis is done with different types of wavelets. The results
computed tomography scans are preferred. CT provides best obtained are suitable for medical image fusion. Smt. G. Ma-
information about denser tissue and MRI offers better infor- matha, L. Gayatri [3] proposed wavelet and curvelet transform
mation on soft tissue [2],[3],[4]. These complementarities have for image fusion and comparison was done between these two
led to idea that combining images acquired with different transform. The results show superiority of curvelet transform
medical devices will generate an image that can offer more than wavelet transform. A. Soma Sekhar, Dr. M. N. Giri Pra-
information than individual image. So, it is expected that fu- sad [4] proposed integration of wavelet transform and PCA
sion of MRI and CT images of the same organ would result in based image fusion for medical images and compare results
an integrated image of much more details [3]. Image fusion is with different types of wavelets, which gives better fusion
deeply related to many different image processing fields such results. Bharat and E.S Karthik Kumar [5] proposed imple-
as satellite imaging, remote sensing and medical imaging. The mentation of 2-G curvelet transform for image fusion which is
study in the field of image fusion has evolved to serve the ad- fast and simpler than 1-G curvelet transform. Researchers
vance in satellite imaging and then, it has been extended to have made lot of work on wavelet and curvelet transform for
the field of medical imaging. Several fusion algorithms have image denoising, image contrast enhancement, fusion of satel-
been proposed extending from the simple averaging to the lite images, image retrieval, texture analysis and object recog-
curvelet transform [1],[3]. nition. Few attempts were made for fusion of the MR and CT
images using curvelet transform. This paper focuses more on
image fusion for medical images. It was found that more work
is done using wavelet transform for fusion of MRI and CT ap-
————————————————
plication. The traditional wavelets perform well only at repre-
senting point singularities since they ignore the geometric
1. Anjali A. Pure,ECE dept.,TIEIT, Bhopal, India. Email-
properties of structures and do not exploit the regularity of
pureanjali@gmail.com
edges.
2. Neelesh Gupta,HOD, ECE dept.,TIEIT, Bhopal, India.
The curvelet transform has evolved as a tool for the repre-
Email- Neelesh9826@gmail.com
sentation of curved shapes and regularity of edges. The wave-
3. Meha Shrivastava,Asst. Professor, ECE dept.,TIEIT, Bhopal,
let transform applied to these two images preserve both spec-
In dia.Email- mehakhare@gmail.com
tral and spatial information and gives image details but it has
limited directionality to deal with curved shapes. The solution
to this is curvelet transform which has ability to deal with
IJSER © 2013
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 130
ISSN 2229-5518

curved shapes. the φj, k(x) and Ψj, k(x) are orthonormal, they include the fol-
lowing property.
To fuse MRI and CT images there are many fusion algo- Vj ⊥ Wj (4)
rithms based on Wavelet. Now in this paper, a new fusion These results in a representation of a single image,
method is proposed based on both wavelet and Second gener-
containing multiscale detail information from all component
ation Curvelet Transform. Since using Wavelet Transform
edge information can’t be extracted clearly. In Medical Image images involved. This representation leads to multiple appli-
Fusion the edge information is most important so a new cations ranging from multispectral image fusion to color and
method can be used to fuse these images. It’s a combination of multivalued image enhancement, denoising and segmenta-
Wavelet and Curvelet based algorithm which is used for fu- tion.
sion to extract edge information and the process is simple to
analyze the fusion of the images. The performance of pro- 2. Bi-orthogonal Wavelet
posed method can be compare with different types of wavelets For biorthogonal transform, perfect reconstruction is
used in image fusion.
available. Orthogonal wavelets give orthogonal matrices and
unitary transforms; biorthogonal wavelets give invertible ma-
trices and perfect reconstruction. For biorthogonal wavelet
2 WAVELET THEORY USED IN IMAGE FUSION filter, the Low-pass and high-pass filters do not the same
2.1 Wavelet Theory length. The low-pass filter is always Symmetrical, while high
Wavelets are finite duration oscillatory functions with zero pass filter could be either symmetric or anti-symmetric. The
average value. The irregularity and good localization proper- method allows unusual flexibility in choosing a filter for any
task involving the multiresolution analysis and synthesis.
ties make them better basis for analysis of signals with dis-

IJSER
continuities. Wavelets can be described by using two func-
3. A’trous (Non-orthogonal) Wavelet
tions viz. the scaling function f (t), also known as father
A‘trous is a kind of non–orthogonal wavelet that is
wavelet and the wavelet function or mother wavelet. Mother
different from orthogonal and biorthogonal. It is a stationary
wavelet ψ (t) undergoes translation and scaling operations to
or redundant transform, i.e. decimation is not implemented
give self-similar wavelet families as in (1). during the process of wavelet transform, while the orthogonal
or biorthogonal wavelet can be carried out using either deci-
mation or undecimation mode. The enhancement of the spatial
information often leads to the distortion of the information in
the spectral domain.
Where, a is the scale parameter and b the translation parame-
ter. Practical implementation of wavelet transforms requires
discretization of its translation and scale parameters by tak-
3 EXISTING METHODS
ing,
3.1 Wavelet Transform
The most common form of transform type image fusion algo-
rithms is the wavelet fusion algorithm due to its simplicity
Thus, the wavelet family can be defined as: and its ability to preserve the time and frequency details of the
images to be fused. Wavelet transform fusion is more formally
defined by considering the wavelet transforms of the two reg-
istered input images together with the fusion rule .Then, the
.
inverse wavelet transform is computed, and the fused image is
2.2 Types of Wavelets reconstructed [1], [3], [5], [6] . 2-D DWT is very useful for im-
The wavelets used in image fusion can be classified into age processing because the image data are discrete and the
three categories Orthogonal, Bi-orthogonal and A’trous wave- spatial and spectral resolution is dependent on the frequency.
let. Although these wavelets share some common properties, The DWT has the property that the spatial resolution is small
each wavelet has a unique image decompression and recon- in low-frequency bands but large in high frequency bands.
struction characteristics that lead to different fusion results
[2],[ 6], [13].

1. Orthogonal Wavelet
The dilations and translation of the scaling function
φj, k(x) constitute a basis for Vj, and Similarly Ψj, k(x) for Wj, if
IJSER © 2013
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 131
ISSN 2229-5518

of radon transform of an image. In this curvelet approach, in-

put image is first decomposed into a set of sub bands each of
which is then partitioned into several blocks for ridgelet anal-
ysis. The ridgelet transform is implemented using the Radon
transform and the 1-D wavelet transform. During the ridgelet
transform, one of the processes is the spatial partitioning
which involves overlapping of windows to avoid blocking
effects. It results in a large amount of redundancy. The
Curvelet Co-efficient has taken and depending upon the max-
imum value the fusion between the images takes place. It is
used to extract the edge information, the ridgelet transform is
complicated so the process is very time consuming and the
Fig.1 Block diagram of Discrete Wavelet transform performance was exceeding slow.
The second generation curvelet transform discarded the use of
A schematic diagram of the wavelet fusion algorithm the ridgelet transform, thus reduced the amount of redundan-
of two registered images I 1 (X 1 , X 2 ) and I 2 (X 1 , X 2 ) is depicted in cy in the transform and increased the speed considerably. Two
fig.1. It can be represented by the following equation, fast discrete curvelet transform (FDCT) algorithm were intro-
I(X 1 , X2) =W-1{Ψ[W (I 1 (X 1 , X 2 )), W (I 2 (X 1 , X 2 ))]} (5) duced. The first algorithm is based on unequally-spaced FFT
while the second is based on the wrapping of specially select-
Where W, W-1and ψ are the wavelet transform operator, the
ed Fourier samples. In this paper, we focus on the “wrapping”
inverse wavelet transform operator and the fusion rule, re-
version of the curvelet transform.
spectively. There are several wavelet fusion rules that can be The algorithm can be summarized as follows-

IJSER
used for the selection of wavelet coefficients from the wavelet 1. Apply the 2D FFT and obtain Fourier samples
transforms of the images to be fused. The most frequently
used rule is the maximum frequency rule which selects the
2. For each scale j and angle l form the product
coefficients that have the maximum absolute values. The
wavelet transform concentrates on representing the image in
multi-scale and it is appropriate to represent linear edges. For 3. Wrap this product around the origin and obtain
curved edges, the accuracy of edge localization in the wavelet
transform is low. So, there is a need for an alternative ap-
Where, the range for n1 and n2 is now 0 < n1 < L1,j and 0 < n2
proach which has a high accuracy of curve localization such as
< L2,j (for θ in the range (−π/4, π/4)).
the curvelet transform.
4. Apply the inverse 2D FFT to each ˜ fj,l, hence collecting the
discrete coefficients cD(j, l, k).
3.2 Curvelet Transform
It is clear that this algorithm has computational complexity
Candes and Donoho justifies that ,though wavelet trans-
O (n2 log n).
form exhibits time frequency localization and yields accepta-
ble fused output, the edges and singularities are not well rep- 3.3 Problem Statement
resented. Also it suffers from limited directionality. The point
The wavelet transform concentrates on representing the
singularity is better suited for wavelets in 1 dimensional sig-
image in multi-scale and it is appropriate to represent linear
nals ,but 2 dimensional signals like images have curve or line
edges. For curved edges, the accuracy of edge localization in
singularities where wavelets fails to approximate. Hence to
the wavelet transform is low. Normally, when a wavelet trans-
process images of sparse nature, Fast Discrete curvelet trans-
formation alone is applied the results are not so useful for
form is recommended by the researchers. E. J. Candes and D.
analysis. For medical applications, MRI and CT images plays
L. Donoho put forward curvelet transform theory in 2005[7].
very important role. Since medical images have several objects
The curvelet transform is a multiscale directional transform
and curved shapes. So, there is a need for an alternative ap-
that allows an almost optimal nonadaptive sparse representa-
proach which has a high accuracy of curve localization, it is
tion of objects with edges. Curvelet transform consisted of
expected that the curvelet transform would be better in their
special filtering process and multiscale Ridgelet transform.
fusion.
Curvelet transform is anisotropy, which can represent the con-
tour of image more sparsely and provide more information for
image processing. The curvelet transform has gone through 4 PROPOSED IMAGE FUSION METHOD
two major revisions. The first generation curvelet transform Wavelet transform is useful for objects with point singu-
used a complex series of steps involving the ridgelet analysis larities and analyses the feature of images in detailed, but it
IJSER © 2013
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 132
ISSN 2229-5518

does not provide information about edges clearly. While 4 CONCLUSION

curvelet transform is more useful for the analysis of images In many important imaging applications, images exhibit edges
having curved shape edges. So, in this paper, a new image and discontinuities across curves. In biological imagery, this
fusion technique based on the combination of wavelet and a occurs whenever two organs or tissue structures meet. Espe-
fast discrete curvelet transform is proposed, which describe cially in image fusion the edge preservation is important in
obtaining the complementary details of the input images. As
the curved shapes of images and analyses feature of images
edge representation in curvelet transform is better and wave-
better. In medical image fusion, edges play a very important
let transform gives image details. The combination of wavelet
role. This new fusion technique is used to fuse the MRI and CT and fast discrete curvelet transform applied to medical images
images. These two images contain complementary infor- may give better fusion results useful for diagnosis. This can be
mation. The Flow of proposed image fusion method can be used to give the edge information clearly and speed of compu-
given as follows: tation will be high compare to other methods. The proposed
method may be useful for researchers for further research
work on image fusion.

REFERENCES
[1] Shih-Gu Huang, Wavelet for Image Fusion.
[2] S.Vasuki, S. Gandhimathi, S. Manika VInodhini,’ Comparative
Analysis of Wavelets for Fusion Application’, IJCA, 2012.

IJSER
[3] Smt.G.Mamatha,L.Gayatri,’AN IMAGE FUSION USING WAVE-
LET AND CURVELET TRANSFORMS’, Global Journal of Ad-
vanced Engineering Technologies,Vol1,Issue-2,2012,ISSN:2277-
6370.
[4] A. Soma Sekhar, Dr.M.N.GiriPrasad, ’A Novel Approach of Image
Fusion on MR and CT Images Using Wavelet Transforms, IEEE
2011.
[5] S.Bharath and E.S. Karthik Kumar, Implementation Of Image Fu-
sion Algorithm Using 2gcurvelet Transforms, ISBN 978-1-4675-
2248-9@2012.
[6] Bin Yang and Shutao Li, Multifocus Image Fusion and Restoration
with Sparse Representation, IEEE 2010.
[7] Emmanuel Cand`es, Laurent Demanet, David Donoho and Lexing
Ying; Fast Discrete Curvelet Transforms,Applied and Computa-
tional Mathematics, Caltech Pasadena 2006.
[8] Jianwei Ma and GerlindPlonka, The Curvelet Transform, IEEE
SIGNAL PROCESSING MAGAZINE[118] MARCH 2010.
[9] Gang Hong, Yun Zhang,’The effect of different types of wavelets
on image fusion.
[10] Deepa M, Wavelet and Curvelet Based Threshholding Techniques
for Image Denoising, IJARCSEE vol1 Issue2012,ISSN:227-9043.
[11] E.J.Candes, D.L.Donoho Curvelets: A Surprisingly Effective Non-
adaptive Reprentation for Objects with Edges.
[12] M. Sifuzzaman M.R. Islam,M.Z.Ali, Application Of Wavelet and
its Advantages Compared to Fourier Transform,Journal of Physi-
cal Science,Vol.13,2009,121-134,ISSN:0972-8791.
[13] Y.Kiran Kumar, Comparison of Fusion Techniques Applied to pre-
Fig. 2 Flow of proposed image fusion method
clinical images: Fast Discrete Curvelet Transform using Wrapping
Technique & Wavelet Transform, JATIT2009.
[14] Pao-Yen Lin, An introduction wavelet transforms.

IJSER © 2013
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 133
ISSN 2229-5518

[15] Myungjin Choi, Rae Young Kim, Moon-GyuKim,’The curvelet

transform for image fusion.
[16] Rafael C. Gonzalez and Richard E. Woods, Digital image Pro-
cessing, Third Edition.

IJSER

IJSER © 2013
http://www.ijser.org