2013 IEEE 9th International Conference on Mobile Ad-hoc and Sensor Networks

A Way of Image Fusion Based on Wavelet Transform

Huaxun Zhang Xu Cao

Electronic Engineering College Center of Modern Education Technology
Changchun University Changchun Chinese Medicine University
Changchun, China Changchun, China
[email protected] [email protected]

Abstract—In this paper a way of medical image fusion based adaptive Wiener filtering, etc. These ways are effectual to
on wavelet theory is introduced. Medical image fusion have wipe off Gaussian noise, impulse noise, multiplicative noise,
three steps, they are image processing, image registration and and so on. But for all kinds of noises in image, this paper
image fusion. In the paper, image processing get across multi suggest a wavelet denoising method. The theory of wavelet
resolution characteristics of wavelet to denoise, image denoising on image is:
registration pass the wavelet analysis to gain biggish change
point and receive image edge to achieve quick and nice The description of local characteristics of signal x(t ) is:

superposable, image fusion use disassumble image to different x(t 0   )  f n (t 0   )  A  , n    n  1
frequency subband to save all information to have a perfect (1)
fusion. Simulation experiment proved it has advantages of
simply calculation, fast superposition and perfect fusion in  is Lipschitz exponent of x(t ) at t0
,
f n (t )
is
medical image fusion. It is a direction of medical reacher and x(t0 ) 
clinic iatrology. polynomial of degree n throu-gh , is a sufficiently
small amount.
Keywords-Wavelet Transform Theory, Image Processing,
If equation (2) is tenable on scale a  2 , then
j
0 is the
Image Registration, Image Fusion
point of local modulus maxima of scale a  2 .
j

I. INTRODUCTION WTx (a, )  WTx (a, 0 ), ( 0  , 0  )

(2)
The mainly proposal of medical image fusion is to
is c. Lipschitz exponent  of signal x(t )
improve the readability of the image by processing the WTx (a, 0 )
redundant data in multiple images and increase image WTx (a, 0 )
sharpness by disposing the complementary information and are satisfied with equation (3),
among multi images. The precondition of image fusion is Log 2 WTx (a, )  Log 2 K  j
picture superimposition. It is significative to image fusion (3)
that only pixels of the two image in the same position are K is a constant associated with the wavelet basis. For
corresponding same anatomic structure. The fusion of multi 1
modality medical image is provide more comprehensive and      0, 0,
2 so modulus maxima of wavelet
accurate data for clinical by consolidating the valuable
physiological functions information and the precise d k( j )
coefficient is decrease with the increase of scale a to
anatomical information. The basal way are based on pixel noise. Moreover wavelet transform is a orthogonal
based and characteristics of the image for the moment. The
former is processing point by point and weighted sum, d k( j )
transformation, therefor is still noise by wavelet
fetched size of gray value. It is a simple method but d ( j)
transform. Viz. k
is accordant in the time domain and
inefficient. The later extract the feathers of the image. It is a
complicated mean but perfect impression is gained[2][3]. frequency domain[5][6].
When scale a is biggish, it fits general observation on
II. A MEASURE OF MEDICAL IMAGE FUSION account of wide field of vision and low frequency. When
The research of medical image fusion is a investigation scale a is minor, it suit detail observation owing to narrow
on multi discipline. It is an anlysis to multi images, but view and high frequency. Wavelet denoising signal is to
because of the orientation of sensor, the change of apparatus analyze the wavelet signal, noise is usually contained in
and all kinds of interfering, different degrees distortion and details of having high frequency, thereby threshold is used to
metamorphism lies in images. So pretreatment and image processe the wavelet coefficient of the decomposition signals.
registration are integrant before image fusion. Afterwards rebuild the wavelet signal, the aim of denoising
achieve. The essence of signal denoise is suppressing the
A. Image Preprocessing useless part and recovering the useful part. Image simulation
Image preprocessing is a course of noise reduction. There results are as Figure 1. and Figure 2.
are sliding average filtering, weighted median filter ,

978-0-7695-5159-3/13 $31.00 © 2013 IEEE 498

DOI 10.1109/MSN.2013.103
 Integral wavelet transform is defined:
1 
 t b
(Wh f )(a, b)  f (t ), ha ,b (t )  a 2
 f (t )h(

a
)dt
(9)

1
t b
ha ,b  a 2 h( )
Thereinto a
Equation (10) and (11) is obtained from equation (8):
Figure 1. The noisy image d s d
Ws' f ( x)  f * ( s )( x)  s ( f *  s )( x)
dx dx (10)
d 2
 d 2
Ws'' f ( x)  f * ( s 2 2
s
)( x)  s 2 2 ( f * s )( x)
dx dx (11)
Standard wavelet transform about  ' ( x) and  (x ) is
Ā

become convolution of f (x) and polishing function s


from equation (10) and (11).
Because of scale s is bigger, the convolution of f (x) and
Figure 2. The denoising image
 s (x) wipe off smaller change, only detect biggish change
B. Image registration point, it’s right of low frequence of wavelet decomposition.
Therefore change points under different scale is obtainned to
Image registration is a premiss of multi image fusion and dissimilar value of s , it is edge detection of image. Edge
multi modality image analysis. If different images want to inspect is solving the maximum of gradient vector modulus
inosculate, they must have oppsite relation on position. There to two dimensional signal[4][8]. After that, image
are many ways of image registration based on different registration still need match corresponding point of two
imaging mode and different relation between objects. images through coordinate transformation. Image of edge
Because of limited length, only a mean of extract edge based detection simulation results is Figure 3 and Figure 4.
on wavelet analysis is refered here.The first of edge detection
based on multi-scale is polishing original signal, then inspect
the change point of original signal from one or two
derivative of polishing signal[1].

Suppose  (x) is polishing function, is satisfied with




  ( x)dx  1 lim x   ( x)  0
, (4)
If  (x) can be derivate by two orders, define
d d2
 ' ( x)   ( x )  ' ' ( x)  2  ( x)
dx , dx (5)
Equation (6) is gained from equation (4):
 

 ( x)dx  0  ( x)dx  0
' ''

 ,  (6)
So  ' ( x) and  (x ) is wavelet. The convolution of
Ā

f , g is defined:

Figure 3. Edge detection of pathological change image
f * g ( x)  ( f * g )( x)   f (u) g ( x  u)du
 (7)
The standard wavelet transform of f (x) is defined
equation (8) about  ' ( x) and  (x ) :
Ā

Ws' f ( x)  f * s' ( x) Wsāf ( x)  f *Ās ( x)

, (8)

499
 Figure 5. Fusion image of pathological change image and no lesion image
Figure 4. Edge detection of no lesion images
III. CONCLUSION
C. Image Fusion This paper introduce a way based on wavelet
The aim of image fusion is combination of different decomposition. It use different characters of wavelet theory.
features of multi-images, reserve important message, stand In image processing, it use enactment threshold to wipe off
out itself character. There are many means of Image fusion, noise. In image registration, it use the convolution of f (x)
here a way based on wavelet decomposition is introducted.  (x) wipe off smaller change and receive image edge
Wavelet transformation fusion is orthogonal transformation, and s
it analyze original image to form sub-image LL of low to achieve quick and nice image registration. In image fusion,
frequency, HL of high frequency in row and low frequency it use disassumble image to different frequency subband to
in arrange, HH of high frequency. Then decompose last sub- save all information to have a perfect fusion. Image
image and gain LL2, HL2, LH2 and HH2 in the multilayer simulation experiment proved that medical image processing
on wavelet theory has advantages of simply calculation, fast
decomposition. Suppose g1 ( x, y ) and g 2 ( x, y) is
superposition and perfect fusion
decomposed G1 ( x, y) and G2 ( x, y) , then the coefficient of
fusion image is:
G (i, j ) | G1 (i, j ) | | G2 (i, j ) | REFERENCES
F (i, j )   1
G2 (i, j ) others (12)
[1] Wolfgang Wein, Shellby Brunke, Ali Khanmene. Automatic CT-
Fusion image f ( x, y ) is received through the inverse ultrasound registration for diagnostic imaging and image-guided
wavelet transform to F (i, j ) [7][9]. Image fusion simulation intervention[J]. Medical Image Analysis, 2008,12:577-585.
results is Figure 5. [2] Taylor RH, Stoianovici D. Medical robotics in computer_ intergrated
surgery. IEEE Transactions on Robotics and Auto- mation 2003,19(5):
765-781.
[3] Vickers JP, Burnham KJ, Dil A.Knowledge-based organ segmen-
hation in low constrast medical computed to mography images [A].
Proceedings of the International Comference on System Science[C],
2001(3): 381-388.
[4] Wyawahare M V, Patil P M, Abhyankar H K. Image Registration
Techniques: An overview[J]. International Journal of Signal
Processing, Image Processing and Pattern Recognition, 2009, 2(3)11-
28.
[5] Mallat S G, Hwang W Lsingularity detection and processing with
waelet[J]. IEEE Trans on Information Theory, 1992, 38(2): 617- 643.
[6] Polchlopek H M, Noonan J P. Wavelet, detection,destination and
sparsity[J]. Digital Signal Processing, 1997(7): 2836-2847.
[7] VandenElsen P A, Pol E J D, Vierge M A. Medical Image matching:
A review with classification[J]. IEEE Eng.Med. Biol.,2009,I2(3):12-
39

500
[8] Stefan Klein, Marius Staring. Evaluation of optimization methods for
nonrigid medical image registration Using Mutual Information and B-
Splinbes[J]. IEEE TRANSACTIONS ON IMAGE PROCESSING,
2007,16(12): 2879-2890.
[9] M. Brown, D.G.Lowe, Recongizing Panoramas. Procedings of IEEE
International Conference on Computer Vision, 2003(2): 1218-1225.
[10] A.Rav-Acha,Y. Pritch, D.Lischinski, et,al. Dynamosaicing:
Mosaicing of Dynamic Scenes, IEEE Trans.PAMI, Oct. 2007: 1789-
1801.

501