International Journal of Advances in Science and Technology,

Vol. 3, No.6, 2011

An Improved Hybrid Model for Medical Image Denoising using Quantitative
Approach

J.Umamaheswari
1
and Dr.G.Rdahamani
2
1
Research Scholar, Dr. G.R.D College of Science, Tamilnadu, India
[email protected]

2
Director, Dr. G.R.D College of Science,Tamilnadu, India. +919443715656

[email protected]

Abstract

This paper provides a new model based on the hybridization of wavelet and relaxed median filter
for denoising of noisy medical images. The present study focuses on proposing a technique to reduce
Gaussian and impulse noise from CT (Computed Tomography) scan devices. In diagnosis of
diseases, devices are frequently used by healthcare professionals. The main problem during
diagnosis is the distortion of visual image signals that are obtained, which is due to the consequence
of the gaussian noise and impulse noise added during transmission or present naturally. We validate
the new model by evaluating the standard brain images in terms of Peak Signal to Noise Ratio
(PSNR), Mean Square Error (MSE) and Elapsed Time (ET). The proposed filter is compared with
existing filters. Experimental results prove the proposed method is efficient.


Keywords: Image denoising, gaussian noise, impulse noise, Wavelet denoising, wavelet
thresholding and relaxed median filter.

1. Introduction

The CT scanner produces an abundant image, which are used by medical practitioners in the
process of diagnosis. The major problem during diagnostic medical images is to reduce Gaussian noise
which is commonly found in medical images and make better image quality. In recent years,
technological development has significantly improved analyzing medical imaging. During
transmission, the capturing devices itself has a impulse noise. These noises corrupt the image and often
lead to incorrect diagnosis. Gaussian is a additive noise, which degrades image quality with a
backscattered wave appearance which originates from many microscopic diffused reflections that
passing through internal organs and makes it more difficult for the observer to discriminate fine detail
of the images in diagnostic examinations. Thus, denoising these Gaussian and impulse from a noisy
image is becoming the most important step in medical image processing.
In image processing applications, linear filters tend to blur the edges and do not remove Gaussian
and mixed Gaussian impulse noise effectively. Previously, a number of schemes have been proposed
for Gaussian mitigation. Inherently noise removal from image introduces blurring in many cases.
Among the two methods, any one method can be used to remove the Gaussian noise, but they should
preserve radiometric information, edge information and spatial resolution. These conditions are met by
Gaussian noise reduction technique.
Many methods have been available for Gaussian denoising [4], [5], [6], [9]. The existing filter used
for Gaussian noise reduction techniques is wavelet filters. Nowadays, the uses of wavelet based
denoising techniques have gained more attention by researchers [1], [2].
Recently a lot of developments have been made use of wavelet transforms not only in image
processing but also in various fields of signal processing. Consequently, wavelet transform is a
powerful tool for modelling non-stationary signals that exhibit slow temporal variations in low
frequency and abrupt temporal changes in high frequency. Hence Gaussian noise can be removed even
in the blurred image.

December Issue Page 27 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology


In this work a fusion technique is proposed to find the best possible solution, so that PSNR, MSE
and ET value of the image after denoising are optimal. The proposed model is based on wavelet
transform and relaxed median filtering, which exploits the potential features of the combination of both
wavelet and relaxed filter denoising.
The paper is organized as follows. In Section 2, we introduce the concept of denoising the medical
image using wavelet transform that works. In Section 3, we explain the work of relaxed median filter.
In Section 4, we propose the fusion model for denoising. In Section 5, we discuss the proposed work
with experimental results applying on standard brain images. Finally we conclude the paper.

2. Wavelet Denoising Model

The wavelet denoising process involves three basic steps as follows:
i) A linear wavelet transforms wavelet coefficients are obtained by applying the linear forward
wavelet transform
ii) Soft thresholding is applied to extract the image form the coefficients of noisy one.
iii) A clear image is obtained after applying linear inverse wavelet transforms.















Figure1. Denoising using wavelets transform filtering

Wavelet Representation of Image

Let f = {f
ij
, i, j = 1, 2M} denote the M N matrix of the original image. During transmission the
image f is corrupted by white Gaussian noise with independent and identically distributed by mean, and
standard deviation. So, the noisy image received at the receiver end is g
ij
= f
ij
+ sn
ij
. The goal is to
estimate the image signal f from noisy observations g
ij
such that Mean Squared Error (MSE) is
minimum and Peak Signal to Noise Ratio (PSNR) is maximum as well as Elapsed Time (ET) should be
minimum within the range [0, 1]. Let W and W
-1
denote the two dimensional orthogonal discrete
wavelet transform (DWT) matrix and its inverse respectively. Then Y = W
*
g represents the matrix of
wavelet coefficients of g having four subbands (LL, LH, HL and HH). The sub-bands HH
k
, HL
k
, and
LH
k
are called details, where k is the scale varying from 1, 2 J and J is the total number of
decompositions. The size of the subband at scale k is M/2
k
N/2
k
. The subband LL
j
is the low-
resolution residue. The wavelet thresholding denoising method processes each coefficient of Y from the
detail subbands with a soft threshold function to obtain X. The denoised estimate is inverse transformed
to f= W.
-1X
These methods use a threshold and determine the wavelet coefficients based on this threshold. There
are two main ways of thresholding the wavelet coefficients, namely the hard thresholding method and
the soft thresholding method.


Noisy
Image

DWT
Wavelet
threshold

IDWT
Denosied
Image
Calculate
Threshold
Apply Soft
Threshold
December Issue Page 28 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology


Hard Thresholding Method

If the absolute value of a coefficient is less than a threshold, then it is assumed to be 0, otherwise it
is unchanged. Mathematically it is

( )( *( ( ) )) X sign Y Y abs Y = >
Where Y represents the noisy coefficients, is the threshold, X represents the estimated
coefficients.

Soft Thresholding Method

Hard thresholding is discontinuous. This causes ringing / Gibbs effect in the denoised image. To
overcome this, Donoho [8] introduced the soft thresholding method. If the absolute value of a
coefficient is less than a threshold, then is assumed to be 0, otherwise its value is shrunk by threshold.
Mathematically it is

( )*(( ( ) )*( ( ) )) X sign Y abs Y abs Y = >
This removes the discontinuity, but degrades all the other coefficients which tend to blur the image.


3. Relaxed Median Filter

An important statistical property of the relaxed median filter is the probability that the center sample
of the windowed sequence be the output; in other words, the probability that a sample remains
unchanged by the relaxed median filter. This probability gives some information about the filter
capacity to preserve details. The performance of the relaxed median filter should be described by using
some statistics of the output. However, this is not possible in general since accurate statistical
descriptions of input images are difficult to obtain. But it is still possible to obtain the probability
distribution function of the output by making simple assumptions about the original image (before
being corrupted by noise). In this sense, the noise attenuation can be well assessed from homogeneous
originals and the detail preservation can be accessed from pure edge and lines.

The output of the relaxed median (RM) filter with parameters and is given by

| | | |
,
( ) ,
,
, [ ]
i
i
i i i i
i otherwise
if
x x w w
y
w RM
x
c

c
= =

e


Where and c are such that 1 2 1 N s s + . Note when 1 = and 2 1 N c = + , the RM, the
identity filter 1 = and 1 N c = + the output of Relaxed median
RM c
. Fig.1 illustrates the
structure of the relaxed median filter. It is worth noting that the symmetric relaxed median filter is also
called rank conditioned median filter [11]. In [12], the output distribution of the relaxed median filter is
given in a more generalized form, and its statistical properties for details preservation are also
presented and illustrated.


December Issue Page 29 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology


Properties of Relaxed median filter
In this subsection, some deterministic properties of relaxed median filters are presented. First,
it is easy to show that relaxed median filters are translation and scale invariant that is

( ) ( ) , ,
s i i
s s t t R
w w RM RM u u
= + e

For the sake of simplicity, we consider a symmetric relaxed median filter, (ie), the parameters and
are symmetric, this restriction is appropriate when the signal and noise are symmetric = 2N + 2
l and to ensure the unbiased filter. Therefore, the symmetric relaxed median RM (2N +1; ) filter is
completely specified by two parameters: the window width and the lower bound . The deterministic
behavior of the relaxed median filter can be analyzed by considering its effect on arbitrary sequences.

4. Proposed Denoised Model

In the present work, two techniques namely, wavelet and relaxed median filters are combined to
form a hybrid denoising model. These techniques are explained below.










Figure 2. Proposed denoising model

The above Figure 2 shows the proposed denoising model. Relaxed median filter can be used in
combination with above equation to remove large spike noises. The proposed hybrid method is defined
as follows:







Noisy
Image

DWT
Wavelet
Threshold

IDWT
Denoised
Image
Relaxed
Median Filter
Preserved
and noise
free image
I/P
Gaussian
suppressed
O/P
To remove impulse
noise and to preserve
edges.
Clean O/P
1. Noisy input images is taken for analysis
2. Apply DWT to form wavelet decomposition.
3. Apply Wavelet Thresholding (Soft Threshold) to separate the image co-efficient from
noisy one.
4. To reconstruct the image IDWT is applied.
5. The gaussian suppressed output is obtained. But it still contain some impulse noise, so
that, relaxed median filter is applied to remove impulse noise and to preserve the
image.
6. After applying relaxed median filter, the unremoval noise is removed by processing
from the initial stage till the noise is removed.

December Issue Page 30 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology



5. Experimental Results

The averages of ten medical images are taken as benchmark images for evaluating results. The real
datasets are taken from scan center. The test bench taken is MATLAB 7.1 version.
The reconstruction of an image has the dimensions of 256 pixel intensity. The medical images in
this contain a wide variety of subject matters and textures. Most of the images used are brain images
with defect and without defect images. The PSNR value must be high for an medical image, MSE and
ET must be less value for a better filtering algorithm.

a.) Peak Signal to Noise Ratio

The PSNR is defined in logarithmic scale, in db (decibels). It is the ratio of peak signal power to
noise power. Since the MSE represents the noise power and the peak signal power, it is unity in case of
normalized image signal. The image metric PSNR is defined as:



Here, MAX
i
is the maximum possible pixel value of the image. When the pixels are represented
using 8 bits per sample, this is 255. More generally, when samples are represented using linear Pulsed
Coupled Modulation (PCM) with B bits per sample, the typical values for the PSNR in lossy image and
video compression are between 30 and 50 dB.

Table 1: Parametric Evaluation for different denoising filters
SRAD KUAN LEE FROST HMF P HMF
PSNR 14.98 14.6 14.51 13.61 13.32 19.79
MSE 2.11E+03 2.30E+03 2.37E+03 2.89E+03 3.07E+03 695
ET 2.88E+00 4.62E+00 3.37E+00 2.26E+00 5.71E+00 4.09
The above Table 1 gives the parametric evaluation for the different denoisng filter and proposed
filter.
The higher the value, the better the images are. Acceptable values for wireless transmission quality
loss are considered to be about 20 dB to 25 dB. When the two images are identical, the MSE will be
equal to zero, resulting in an infinite PSNR.
The following Figure 3 gives the PSNR value for the proposed filter. Compare to the existing filter,
the proposed filter gives most suitable result.






December Issue Page 31 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology




Figure 3. PSNR value for the proposed filter

b.) Mean Square Error (MSE)
The metric MSE is defined as:

For two mn monochrome images I and K, one of the images is considered a noisy approximation
of the other.



Figure 4. MSE value for the proposed filter
The above Figure 4 shows the MSE value for the proposed filter. Compare to other filter, the
proposed filter gives most suitable result.
December Issue Page 32 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology


c.) Evaluation Time
Evaluation Time (ET) of a filter is defined as the time taken by a digital computing platform to
execute the filtering algorithms when no other software, except the operating system (OS), runs on it.
Though ET depends essentially on the computing systems clock time-period, yet it is not necessarily
dependant on the clock time alone. Rather, in addition to the clock-period, it depends on the memory-
size, the input data size, and the memory access time. However, the measure ET is very important in
case of real-time application.
The execution time taken by a filter should be low for online and real-time image processing
applications. Hence, a filter with lower ET is better than a filter having higher ET value when all other
performance-measures are identical.



Figure 5. ET value for the proposed filter
The above Figure 5 shows the ET value for the proposed filter. Compare to the existing filter, the
proposed filter gives most suitable result compared to the HMF filter.
The experimental results are proved by subjective and objective method. The following Table 1
shows the performance evaluation of fusion technique for denoising medical images. From the table,
the proposed fusion filter gives most suitable results based on PSNR, MSE and ET value.

6. Conclusion

A new model based on the hybridization of wavelet and relaxed median filters for denoising of
noisy medical images is developed. In diagnosis of diseases CT devices are frequently used by
healthcare professionals. The main problem during diagnosis is the distortion of visual image signals
obtained which is due to the consequence of the gaussian noise and also during transmission impulse
noise is also added. The model is experimented on real scan data. The performances are evaluated in
terms of peak signal to noise ratio (PSNR), Mean Square Error (MSE) and Elapsed Time (ET). The
proposed filter is compared with existing filters. Experimental results prove that the proposed method
is efficient in reaching convergence quickly and producing quality based on PSNR, MSE and ET
values.



December Issue Page 33 of 85 ISSN 2229 5216
International Journal of Advances in Science and Technology


7. References
[1] G.Chang, B.Yu, and M.Vetterli, Adaptive Wavelet Thresholding for Image Denoising and
Compression ,IEEE Trans of Image Processing, Vol. 9, No. 9, pp. 1532-1546, 2000.
[2] I.Delakis, O. Hammad, and R.I Kitney, Wavelet-based de-noising algorithm for images
acquired with parallel magnetic resonance imaging (MRI), Physics in Medicine and Biology,
Vol. 52, No.13, pp. 3741, 2007.
[3] A.B.Hamza, P.L.Escamilla, J.M.Aroza, and R.Roldan, Removing Noise and Preserving
Details with Relaxed Median Filters, Journal of Mathematical Imaging and
Vision,Vol.3,No.1, pp. 161-177, 1999.
[4] R.Hossein, V.Mansur, A.Purang, and G.Saeed, Speckle Noise Reduction of Medical
Ultrasound Images in Complex Wavelet Domain Using Mixture Priors, IEEE transactions on
biomedical engineering , Vol. 55, No. 9, pp. 2152-2160, 2008.
[5] J. Rajan, K. Kannan, and M.R.Kaimal, An Improved Hybrid Model for Molecular Image
Denoising, Journal of Mathematical Imaging and Vision, Vol. 31, No.1, pp. 7379, 2008.
[6] Y.Yu, and S.T.Acton, Speckle Reducing Anisotropic Diffusion, IEEE Transactions on
Image Processing, Vol. 11, No. 11, pp. 1260 1270, 2002.
[7] L.Gagnon, A. Jouan, Speckle filtering of SAR images: A comparative study between
complex-wavelet based and standard filters, In SPIE Proc, pp. 8091, 1997.
[8] D. L. Donoho, De-noising by Soft Thresholding, IEEE Trans. on Inform Theory, Vol.41,
No. 3, pp. 613- 627,1995.
[9] L.Qiang, Y.Zhengan, K.Yuanyuan, Solutions of Fourth-order Partial Differential Equations
in a noise removal model, Electronic Journal of Differential Equations, No.120, pp 1-11,
2007.
[10] P. Pietro and M.Jitenda, Scale space and Edge Detection using anisotropic diffusion, IEEE
Transactions on Pattern Analysis and Machine Intelligence, No.7, vol.12, pp 629-639, 1990.
[11] T. Loupas, W. McDicken, and P. Allan, An adaptive weighted median filter for speckle
suppression in medical ultrasound image, IEEE Trans. Circuits Syst., vol. 36, pp. 129-135,
1989.
[12] J. S. Lee, Digital image enhancement and noise filtering by use of local statistics, IEEE
Trans. Pattern Anal. Mach. Intell, vol. 2, pp. 165168, 1980.



Authors Profile



Ms. J.Umamaheswari Research Scholar in Computer Science, Dr. G.R.D college,
Coimbatore. She has 5 years of teaching experience and two years in Research. Her areas of
interest include Image Processing, Multimedia and communication. She has more than 3
publications at International level. She is a life member of professional organization IAENG.







Dr. G. Radhamani Professor & Director in School of Computer Scienec, Dr.G.R..D College,
Coimbatore. She was awarded Bharat Joythi in lieu of the yeomen services rendered with
excellence in their respective fields. She has worked for the Malaysia Venture Capital
Management Ph.d, and has been a reviewer for IEEE Wireless Communications Magazine and
Multimedia Cyberscape Journal. She has published two books, namely, Web Services Security
and e-Business, and WiMAX A Wireless Technology Revolution. Her areas of interest
include Mobile Computing, e-internet and communication. She is a member IEEE.


December Issue Page 34 of 85 ISSN 2229 5216