Anda di halaman 1dari 4

192 (IJCNS) International Journal of Computer and Network Security,

Vol. 2, No. 10, 2010

Denoising of Magnetic Resonance Images using


Wavelets- A Comparative Study
S Satheesh1 Dr.KVSVR Prasad2 P.Vasuda3
1
Asst.Prof., Dept. of ECE, G Narayanamma Institute of Technology and Science, Hyderabad, , India
satheesh.s17@gmail.com
2
Prof. & Head Dept. of ECE, D.M.S.S.V.H. College of Engineering, Machilipatnam, India
3
Asst.prof., Dept. of ECE, G Narayanamma Institute of Technology and Science, Hyderabad, India

Abstract: Image denoising has become an essential exercise in such as hard thresholding, soft thresholding and Wiener
medical imaging especially the Magnetic Resonance Imaging filter is compared both visually and in the PSNR sense.
(MRI). As additive white Gaussian noise (AWGN) exhibits finest
grain property of noise, multi resolution analysis using wavelet 2. Wavelet Based Image Denoising
transform is gaining popularity. The aim of the work is to
compare the effectiveness of three wavelet based denoising DWT has attracted more interest in image denoising [5].
algorithms viz. Wiener filter, hard threshold and soft threshold The DWT can be interpreted as image decomposition in a
using MRI images in the presence of AWGN. Wiener filter set of independent, spatially oriented frequency channels.
performs better visually and in terms of PSNR than the other The image is passed through two complementary filters and
thresholding techniques. emerges as two images, Approximation and Details. This is
Keywords: Denoising, Wavelet, MRI, Wiener filtering, called Decomposition or Analysis. The components can be
Threshold
assembled back into the original image without loss of
information. This process is called Reconstruction or
1. Introduction Synthesis. The mathematical manipulation, which implies
Image denoising is a procedure in digital image analysis and synthesis, is called DWT and inverse DWT.
processing aiming at the removal of noise, which may For a 2D image, an N level decomposition can be performed
corrupt an image during its acquisition or transmission, resulting in 3N+1 different frequency bands namely, LL,
while retaining its quality. Medical images obtained from LH, HL, HH.
MRI are the most common tool for diagnosis in medical Denoising algorithms that use the wavelet transform consist
field. These images are often affected by random noise of three steps:
arising in the image acquisition process. The presence of • Calculate the wavelet transform of the noisy image
noise not only produces undesirable visual quality but also • Modify the noisy wavelet coefficients according to
lowers the visibility of low contrast objects. Noise removal is some rule.
essential in medical imaging applications in order to
• Compute the inverse transform using the modified
enhance and recover anatomical details that may be hidden
in the data. coefficients.
The wavelet transform has recently entered the arena of 2.1 Wiener Filter
image denoising and it has firmly established its stand as a In signal processing, the Wiener filter is a filter proposed
powerful denoising tool. There has been a fair amount of by Norbert Wiener during the 1940s.Its purpose is to reduce
research on filtering and wavelet coefficients thresholding the amount of noise present in a signal by comparison with
[8], because wavelets provide an appropriate basis for an estimation of the desired noiseless signal. The discrete-
separating noisy image from the original image. These time equivalent of Wiener's work was derived independently
wavelet based methods mainly rely on thresholding the by Kolmogorov in 1941. Hence the theory is often called the
discrete wavelet transform (DWT) coefficients, which have Wiener-Kolmogorov filtering theory.
been affected by AWGN. The inverse filtering is a restoration technique for
There has been much research by Donoho & Johnstone [1, deconvolution, i.e., when the image is blurred by a known
2, 3] on finding thresholds, however few are specifically lowpass filter, it is possible to recover the image by inverse
designed for images. One of the most popular method filtering or generalized inverse filtering. However, inverse
consists of thresholding the wavelet coefficient (using the filtering is very sensitive to additive noise. The approach of
hard threshold or the soft threshold) as introduced by reducing one degradation at a time develops a restoration
Donoho. algorithm for each type of degradation and simply combines
Another denoising method in the wavelet domain consists of them. The Wiener filtering executes an optimal tradeoff
Wiener filtering the wavelet coefficients. In this paper, the between inverse filtering and noise smoothing. It removes
performance of this method is done on a degraded image X the additive noise and inverts the blurring simultaneously.
such that X=S+N where S is the original image and N is an
AWGN. The performance of three denoising techniques
(IJCNS) International Journal of Computer and Network Security, 193
Vol. 2, No. 10, 2010
The Wiener filtering is optimal in terms of the mean When using the Haar wavelet transform the steps for
square error. In other words, it minimizes the overall mean implementing denoising using the Wiener filter technique is
square error in the process of inverse filtering and noise as follows:
smoothing. The Wiener filtering is a linear estimation of the i. Apply the Haar wavelet transform to the original image

{ } {y }
original image. The approach is based on a stochastic 2
framework [6, 7]. ii. qi , j is computed by convolving i, j with a
2.1.1 Wiener Filter in the Wavelet Domain kernel of size 9.
In the model we assume that the wavelet coefficients are iii. The Wiener filter is then applied using the formula
conditionally independent Gaussian random variables. The max(qi , j − σ n2 , 0)
noise is also modeled as stationary independent zero-mean sˆ = yi , j = aˆi , j yi , j
qi , j
Gaussian variable. Let us consider an image corrupted by a
zero-mean Gaussian noise. The coefficients of the noisy iv. Apply the inverse Haar wavelet transform.
image in the wavelet domain are given by [4].
yi, j = si, j + ni , j (1) 2.2 Soft Thresholding
Where yi , j represent the coefficients of the noisy image in In soft thresholding, the wavelet coefficients with
magnitudes smaller than the threshold are set to zero, but
the wavelet domain, si , j represent the coefficients of the the retained coefficients are also shrunk towards zero by the
amount of the threshold value in order to decrease the effect
undegraded image, ni , j represent the coefficients of the
of noise assumed to corrupt all the wavelet coefficients. Soft
noise. thresholding shrinks the coefficients above the threshold in
Without loss of generality, we can assume that absolute value.
the E { yi2, j } ’s can be determined by averaging the squared When using the Haar wavelet transform, the steps for
implementing denoising using the soft thresholding
values of yi , j in a window centered at (i, j). This technique is as follows:
information can be expressed as • Apply the Haar wavelet transform to the original
R R image
Qi , j = ∑ ∑y
k =− R l =− R
2
i − k , j −1 • Apply the soft thresholding on the wavelet
(2) coefficients y(i, j )
M = (2R + 1)2
(3)
 y − T if y(i , j ) ≥ T
Qi , j 
qi , j = s$ (i, j ) =  y + T if y(i , j ) ≤ −T
M  0
(4)  otherwise
As a result, the coefficients of the Wiener filter can be (9)
expressed as T = σ 2 log ( n )
(10)
qi , j − σ n2 Where y(i, j ) is the standard deviation of the noise, n is
ai, j =
qi , j the number of wavelet coefficients, sˆ(i, j ) are the de-
(5) noised wavelet coefficients, and T is the universal
Restricting the values to only positive values, the numerator threshold and the variance is estimated using MAD
of the equation (4) takes the form max ( qi , j − σ n2 , 0 ) and so method.
sˆi , j = max ( a i , j , 0 ) yi , j • Apply the inverse haar wavelet transform.
(6) 2.3 Hard Thresholding
Where sˆi , j is the best linear estimate of the signal
In hard thresholding, the wavelet coefficients with greater
component si , j magnitudes than the threshold are retained unmodified as
The noise variance is estimated using the mean absolute they are thought to comprise the informative part of data,
deviation (MAD) method and is given by while the rest of the coefficients are considered to represent
2 noise and set to zero. However, it is reasonable to assume
 mad 
var iance =   (7) that coefficients are not purely either noise or informative
 0.6745  but mixtures of those.
mad ( ( wi )i ) = median (( w ) )
i i
The denoising method described in the previous
subsection (soft thresolding) can be carried out using the
(8) hard threshold instead of the soft threshold on the wavelet
wi represents the wavelet coefficients. coefficients in (ii).
The hard thresholding formula is given as
194 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 10, 2010
pretty well in the smooth regions but performs poorly along
f y(i , j ) ≥ T the edges.
 y(i, j )
sˆ(i, j ) = 
0 if y(i, j ) < T The comparison of PSNR of the three wavelet filters for
(11) different MRI images are tabulated in Table1 and is
observed that the Wiener filter gives better values compared
3. Results and Discussion to soft and hard thresholding for different noise variances
(σ ) such as 15, 20, 25, 30.
In this section, simulation results are presented which is
performed on the four MRI images i.e. Brain, Knee, Spine
Abdomen. White Gaussian noise is added to the MRI Table 1: Comparison of PSNR of different wavelet filters
images and denoised with the methods described previously. for different MRI images corrupted by AWGN
The performance of the three techniques is compared
using PSNR , which is defined as
255 2
PSNR = 10log10
MSE Peak Signal to Noise Ratio in dB(PSNR)
Image Noise
(12)
(σ )
Where MSE denotes the mean square error for two
Hard Soft Wiener
m × n images l (i , j ) & k(i, j ) where one of the images is
Thresholding Thresholding Filter
considered a noisy approximation of the other and is given
15 25.441 26.946 27.692
as
2
Brain 20 22.936 25.065 25.615
1 m−1 n −1
MSE = ∑∑ [l (i, j ) − k(i, j )]
mn i =0 j = 0
25
30
21.031
19.452
23.507
22.181
23.957
22.577
(13) 15 25.219 26.231 26.803
From the simulation results it has been observed that the Knee 20 22.810 24.542 24.977
Wiener filter outperforms both thresholding methods 25 20.924 23.087 23.457
visually and in terms of PSNR. More details were lost with 30 19.357 21.836 22.170
the thresholding methods especially for the hard threshold 15 25.093 26.069 26.561
Spine 20 22.753 24.458 24.812
wherein the background was not well denoised. If the
25 20.869 23.073 23.359
Wiener filter could be thought as another thresholding
30 19.348 21.871 22.126
function, it will perform better as its shape is smoother than 15 25.402 27.135 27.656
the hard and soft thresholds. Abdomen 20 22.917 25.227 25.614
This can be clearly seen from Figure1 and Figure2 that the 25 21.011 23.660 23.981
background of the denoised images with Wiener filter 30 19.436 22.324 22.622
appears smoother. The Wiener filter removes the noise

Figure1. Denoising of Brain MRI image for variance=20 (a) Original image (b)Noisy image (c)Denoised image with hard
threshold (d) Denoised image with soft threshold (e) Denoised image with Wiener filter

(a) (b) (c) (d) (e)


Figure2. Denoising of Spine MRI image for variance=30 (a) Original image (b)Noisy image (c)Denoised image with hard
threshold (d) Denoised image with soft threshold (e) Denoised image with Wiener filter
(IJCNS) International Journal of Computer and Network Security, 195
Vol. 2, No. 10, 2010

4. Conclusions Author’s Profile


The paper presents a comparative analysis of three image
denoising techniques using wavelet transforms. The analysis S.Satheesh received the B.Tech degree in
of all the experimental results demonstrates that Wiener Electronics and Communication
filter surpasses other methods that have been discussed. Engineering from VRSEC (ANU,
Vijayawada, India) in 2001, and the M.E
There are a couple of areas which would like to be improved
(ECE) degree in Communication
on. One area is in improving the denoising along the edges Engineering specialization from CBIT
as the Wiener method did not perform so well along the (OU, Hyderabad, India) in 2005.He is
edges. Another area of improvement would be to develop a currently pursuing the Ph.D. degree under
better optimality criterion as the MSE is not always the best the guidance of Dr. KVSVR Prasad at
optimality criterion. Jawaharlal Nehru Technological
University Hyderabad, India. He is the member of ISTE, IAENG
Acknowledgements and IACSIT. His research interests are in the area of Medical
Image Processing and Signal Processing.
We wish to express our sincere thanks to Dr. K. Jitender
Reddy, Consultant Radiologist, Dept. of Radiology & Dr. KVSVR Prasad obtained B Sc.
Imaging Sciences, Apollo Health City, Hyderabad for Degree in 1963 from Andhra University,
providing us with different MRI image datasets. B.E (Telecommunication Engineering) in
1967 and M.E (ECE) Microwave
Engineering specialization in 1977 from
References Osmania University. He received the
[1] DonohoD.L and. Johnstone I.M “Ideal spatial Ph.D. in 1985 from IIT Kharagpur in strip
adaptation via wavelet shrinkage” Biometrica, Vol. and micro strip transmission lines. He
published six papers in IEEE Transactions
81, pp. 425-455,1994
in MTT, Antenna and Propagation and
[2] Donoho D.L “De-noising by soft-thresholding” IEEE EMI/EMC and three papers in National
Transactions on Information Theory, Volume: 41, Conferences. He is fellow of IETE (life member). He worked in
Issue: 3, Pages: 613 – 627, May 1995 various capacities in the Department of ECE, Osmania University,
Hyderabad. Presently he is working as professor and head,
[3] Donoho, D.L “Wavelet Shrinkage and W.V.D.: A 10-
Department of ECE, D.M.S.S.V.H. college of engineering,
minute Tour; (David L. Donoho's website) Machilipatnam, India.
[4] Kazubek, M “Wavelet domain image denoising by
thresholding and Wiener filtering” IEEE Signal P.Vasuda received B.Tech degree from
Processing Letters, Volume: 10, Issue: 11, Nov. 2003 G.Narayanamma Institute of Technology
and Science, Hyderabad, India in
[5] Kother Mohideen S, Dr. Arumuga Perumal. S , Dr. Electronics and Communication
Mohammed Sathik M “Image Denoising using Discrete Engineering in the year 2007.Currently
Wavelet Transform”, International Journal of Computer she is pursuing her M.Tech degree from
Science and Network Security, Vol.8, No.1, January G.Narayanamma Institute of Technology
2008. and Science, Hyderabad, India in Digital
[6] Lakhwinder Kaur , Savita Gupta , R.C. Chauhan Electronics and Communication
Engineering. Her area of interest includes
“Image Denoising using Wavelet Thresholding” Third
Image Processing and Digital Communications.
Conference on Computer Vision, Graphics and Image
Processing, India, Dec 16-18, 2002
[7] Nevine Jacob and Aline Martin “Image Denoising in
the Wavelet domain using Wiener filtering”, December
17, 2004.
[8] Zhong, S, Cherkassky V “Image Denoising using
Wavelet Thresholding and Model Selection” Image
Processing 2000, Proceedings. 2000 International
Conference on, Volume: 3, Pages: 262 -265, 10-13
Sept. 2000

Anda mungkin juga menyukai