Versatile Medical Image Denoising Algorithm

Indian Streams Research Journal
Vol -3 , ISSUE 4, May.2013

ISSN:-2230-7850
Available online at www.isrj.net
ORIGINAL ARTICLE
AN EFFICIENT MEDICAL IMAGE DENOISING ALGORITHM
SRINIVASAKIRAN GOTTAPU AND M.VENUGOPAL RAO
Dept Of ECE , K.L University ,Andhra Pradesh , India.

Professor ,Dept Of ECE ,K.L University Andhra Pradesh , India.
ABSTRACT
Noise suppression in medical images is a particularly delicate and difficult task. A trade-off
between noise reduction and the preservation of actual image features has to be made in a way that
enhances the diagnostically relevant image content. Image processing specialists usually lack the
biomedical expertise to judge the diagnostic relevance of the De-noising results. For example, in
ultrasound images, speckle noise may contain information useful to medical experts the use of speckled
texture for a diagnosis was discussed in. Also biomedical images show extreme variability and it is
necessary to operate on a case by case basis. This motivates the construction of robust and Efficient
denoising methods that are applicable to various circumstances, rather than being optimal under very
specific conditions. In this paper, we propose one robust method that adapts itself to various types of image
noise as well as to the preference of the medical expert: a single parameter can be used to balance the
preservation of relevant details against the degree of noise reduction. The proposed algorithm is simple to
implement and fast. We demonstrate its usefulness for denoising and enhancement of the CT, Ultrasound
and Magnetic Resonance images.

ISSN:-2230-7850
Index Terms
Medical Image denoising, Efficient Noise Filtering, Noise Suppression.
I.
INTRODUCTION
Fourier and related representations are classical methods that have been widely used in image
processing applications. The noise removal has been done using the Wiener filter, which is derived by
assuming a signal model of uncorrelated Gaussian distributed coefficients in the Fourier domain and
utilizes second-order statistics of the Fourier coefficients. Statistical modelling of image features at
multiple resolution scales is a topic of tremendous interest for numerous disciplines including image
restoration, image analysis and segmentation, data fusion, etc.
Noise suppression in medical images is a particularly delicate and difficult task. A trade-off
between noise reduction and the preservation of actual image features has to be made in a way that
enhances the diagnostically relevant image content. Image processing specialists usually lack the
biomedical expertise to judge the diagnostic relevance of the De-noising results. For example, in ultrasound
images, speckle noise may contain information useful to medical experts; the use of speckled texture for a
diagnosis was discussed in. Also biomedical images show extreme variability and it is necessary to operate
on a case by case basis. This motivates the construction of robust and Efficient De-noising methods that are
applicable to various circumstances, rather than being optimal under very specific conditions.
The notion of robustness in multi-scale denoising was addressed in. In this Paper, we propose one
robust method that adapts itself to various types of image noise as well as to the preference of the medical
expert: a single parameter can be used to balance the preservation of relevant details against the degree of
noise reduction. Lee proposed digital image enhancement and noise filtering by use of local statistics.
Recently Pizurica and co-authors [9] proposed a low-complexity joint detection and estimation method. In
particular, the method applies the minimum mean squared error criterion assuming that each wavelet
coefficient [10][11]represents a signal of interest with a probability leading to the generalized likelihood

ISSN:-2230-7850
ratio formulation in the wavelet domain. They proposed an analytical model for the probability of signal
presence, which is adapted to the global coefficient histogram and to a local indicator of spatial
activity(e.g., the locally averaged magnitude of the wavelet coefficients).
In this paper, we propose a related, but more flexible method, which is applicable to various and
unknown types of image noise. We employ a preliminary detection of the wavelet coefficients that
represent the features of interest in order to empirically estimate the conditional pdfs of the coefficients
given the useful features and given background noise. At the same time, the preliminary coefficient
classification is also exploited to empirically estimate the corresponding conditional pdfs of the local
spatial activity indicator (LSAI). The preliminary classification step in the proposed method relies on the
persistence of useful wavelet coefficients across the scales, and is related to the one in, but avoids its
iterative procedure.
The classification step of the proposed method involves an adjustable parameter that is related to the
notion of the expert-defined relevant image features. In certain applications the optimal value of this
parameter can be selected as the one that maximizes the signal-to-noise ratio (SNR) and the algorithm can
operate as fully automatic. However, we believe that in most medical applications[10] the tuning of this
parameter leading to gradual noise suppression[11] may be advantageous. The proposed algorithm is
simple to implement and fast. We demonstrate its usefulness for denoising and[12] enhancement of the CT,
Ultrasound and Magnetic Resonance images.
This paper is organized as follows. Section-II describes the basic theory of the proposed algorithm.
Section-III illustrate the versatile Noise filtration technique. Section IV
elaborate the Summary of
proposed algorithm. Section V and VI shows the Implementation of proposed algorithm to Ultra sound
images and magnetic resonance images respectively, Results are discussed and conclusion is given in
section VII.

ISSN:-2230-7850
II.
BASIC THEORY OF PROPOSED ALGORITHM
In this section we will study the theoretical concept behind the proposed method, and then the new
practical algorithm is described. We start from a general noise model y k = w k nk , where wk is the
unknown noise-free wavelet coefficient, a point-wise mathematical operation (addition in the case of
additive noise and multiplication in the case of speckle noise) and nk an arbitrary noise contribution. Our
wavelet domain estimation approach relies on the joint detection and estimation theory and is related to
the problem of the spectral amplitude estimation. The algorithm is implemented using the quadratic
spline wavelets
Let X k denote a random variable, which takes values xk from the binary label set {0,1} . The
hypothesis the wavelet coefficient y k represents a signal of interest is equivalent to the event X k = 1 ,
and the opposite hypothesis is equivalent to X k = 0 . The wavelet coefficients representing the signal of
interest in a given sub band are identically distributed random variables with the probability density
function p Yk | X k ( w k |1) . Similarly, the coefficients in the same sub band, corresponding to the absence of
the signal of interest, are random variables with the pdf pY
|X k
(w k | 0) .
Under the model assumptions, the minimum mean squared error estimate (the conditional mean) of w k
is w k = E (wk | yk , X k = 1) P ( X k = 1 | yk ) + E (wk | yk , X k = 0) P ( X k = 0 | yk ) where E (.) stands for the expected
value. If the signal of interest is surely absent in a given wavelet coefficient, then Wk0and
E(wk|yk,Xk=0)0. In the case where the signal of interest is surely present, we approximate
E(wk|yk,Xk=1)yk ,which accounts for the fact that vast majority of the coefficient magnitudes representing

ISSN:-2230-7850
the signal of interest are highly above the noise level. Applying Baye's rule, one can express P ( X k = 1 | yk ) as
a generalized likelihood ratio, and our estimate becomes
w k =
Where k =
p Y | X ( yk | 1)
k
p Y | X ( yk | 0)
k
k k
1+ k k
k =
yk
P ( Xk = 1 | P )
(2)
P ( Xk = 0 | P )
And P symbolically denotes the prior knowledge that is used to estimate the probability of signal
presence.
Pizurica[1][3][9] proposed a method to estimate this probability for each wavelet coefficient from
its local surrounding, using a chosen indicator ek of the local spatial activity. In particular, since our
estimate of the probability of signal presence is a function of ek , we write
P ( X k = 1 | P ) = P ( X k = 1 | ek ) , and replace
k by
k =
(
) = r p Yk | Xk (ek | 1)
p Y | X (ek | 0)
P ( Xk = 0 | ek )
k k
P Xk = 1 | ek
(3)
Whereris the ratio of unconditional prior probabilities
r=
P ( X k = 1)
(4)
P ( X k = 0)
For a given type of noise, one can derive the complete estimator analytically. In such approaches
where the required conditional densities need to be expressed analytically, the choice of the local spatial
activity indicator is usually restricted to simple forms: even when ekis defined simply as the locally
averaged coefficient magnitude, certain simplifying assumptions about the statistical properties of the

ISSN:-2230-7850
wavelet coefficients are needed in order to derive pE | X (ek | xk ) analytically. The algorithm that we
k k
propose in this paper is applicable to various noise types, and allows an arbitrary choice of ek.
III.
THE EFFICIENT NOISE FILTERING TECHNIQUE
The idea behind the proposed algorithm [4][6] is to empirically estimate the probabilities and the
probability density functions that specify the estimator. Let N denote the number of wavelet coefficients
in a detail image. For each detail image Y jD = { y D1, j , . . , y DN , j } , we first estimate the mask
D = x D , . . ., x D
X
j
N, j
1, j
} which
indicates the positions of significant wavelet coefficients (representing the
signal of interest). As usual, we relate the notion of significant wavelet coefficients to the standard
deviation of the noise. Also, we rely on the persistence of significant wavelet coefficients across resolution
scales. In particular, we extend our robust coarse-to-fine classification method from [1][9] as follows:
0, if | y Dk , j || w Dk , j | < K ( Dj ) 2
x kD, j =
D
D
D 2
1, if | y k , j || w k , j | K ( j )
Where Dj an estimate of the noise standard deviation in the detail is image YjD , and K is a heuristic,
tunable parameter that controls the notion of the signal of interest. We estimate the standard deviation
of the input noise as the median absolute deviation of the wavelet coefficients in the HH sub band at
the finest resolution scale, divided by 0.6745.
In estimating Dj , we follow ( Dj ) 2 = S Dj 2 , where for each sub band the constant S Dj is calculated
from the filter coefficients of the high pass filter g and the low pass filter h of the discrete wavelet

ISSN:-2230-7850
transform, as S jLH , HL = k g
2
k
)(
l h
2
l
) and S
HH
j
) ( l h2l )
22
= k gk
2( j 1)
[8]. To initialize the classification, we
D = Y D , where J is the coarsest resolution level in the wavelet decomposition.

start from W
J
J
Now we address the estimation of the wavelet coefficients YJD using the estimated mask x Dj . The
estimator requires the conditional densities p Y | X ( y k | xk ) and p E | X (e k | x k ) . Since p Y | X ( y k | xk ) is usually
k k
k k
k k
highly symmetrical around 0 , in practice we shall rather estimate the conditional pdfs pM | X ( mk | xk ) of
k k
the coefficient magnitudes mk =| yk | . As the local spatial activity indicator ek , we use the averaged energy
of the neighboring coefficients of yk where the neighbors are the surrounding coefficients in a square
window at the same scale and the parent (i.e., the coefficient at the same spatial position at the first
coarser scale). Having the estimated mask x = { x1..x N } , Let S0 = {k : xk = 0} , and S1 = {k : xk = 1} . The empirical
estimates p M
k | Xk
( mk | 0 )
and p E | X (ek
k
| 0)
are computed from the histograms of {mk : k S0 } and {ek : k S0 }
respectively (by normalizing the area under the histogram). Similarly, p M | X ( yk | 1) and p E | X (ek | 1) are
k k
k k
computed from the corresponding histograms for k S1 .
Our estimation approach still requires the probability ratio. Reasoning that P ( X k = 1) can be
estimated as the fractional number of labels for which x k = 1 , we estimate the parameter r from previous
equation as
r =
kN=1 xk
N kN=1 xk
Then the final estimation is defined as

ISSN:-2230-7850
w k =
r kk
yk
1 + r kk
Where
k =
p M | X ( mk | 1)
k k
p M | X ( mk | 0)
k k
And k =
p E | X (ek | 1)
k k
p E | X (ek | 0)
k
In Fig. 1, we show an example of the empirical densities p M | X ( mk | xk ) and p E | X (ek | xk ) . The

k k
k k
direct computation of the ratios k and k from the normalized histograms shown in Fig. is not appropriate
due to errors in the tails. One solution is to first fit a certain distribution to the histogram. Here we apply a
simpler approach, observing that both log (k ) and log (k ) can be approximated well by fitting a piecewise linear curve as illustrated in Fig. 1. Formally, we approximate
log( k ) =
a1 + b1mk , k < 1
a + b m , 1
2 2 k k
c1 + d1mk , k < 1
log( k ) =
c + d m , k 1
2 2 k
The
fitting
in
is
done
as
follows.
For
R = {l : p M | X (m l | 0) 0, 0, pM |X (ml |1) 0} we
{ (
log = log p M | X ( m l
| 1)
the
index
set
compute
) log ( p M | X (m l | 0) ) : l R } Further,
let
R = {l : pM |X ( ml | 0)
the
R ,1 = l : l R , log, l < 0
} . We look for the polynomial coefficients a1, b1 , which minimize
R ,2 = l : l R , log, l 0
log-ratios
and

ISSN:-2230-7850
lR ,1 a 1m l + b 1 log,l
),
2
And the polynomial coefficients a2 , b2 , which minimize
lR ,2 a 2 m l + b 2 log,l
i.e., which are optimal in the least square sense. The equivalent procedure is applied to find the
coefficients c1 , d1 and c2 , d 2 . This completes the specification of the proposed method.
IV.
SUMMARY OF PROPOSED ALGORITHM
Step 1. Compute the non-decimated wavelet transform with J resolution levels.

Step 2.Initialize W DJ = Y DJ ,
D {H L, LH , H H } .
Step 3. For each orientation D and for each scale

2 j,
j = 1, ..., J 1 ,
and
(a) For all the spatial positions k = 1, ..., N .

(i) Estimate x D
0 ,
1,
k ,j =
2
if | y Dk , j || w Dk , j | < K ( D
j )
2
i f | y Dk , j || w Dk , j | K ( D
j )
(ii) Compute the local spatial activity indicator
ek
using averaged energy of the neighboring coefficients of
yk where the neighbors are the surrounding coefficients in a square window.
(b) Compute. r = lN=1 xl / N lN=1 xl
(c) Define S0 = {l : xl = 0} and S1 = {l : xl = 1} .

Estimate p M | X ( m | 0) p M | X ( m | 1) p M | X ( e | 0) and p M | X ( e | 1) , from the
Corresponding histograms of ml =| Wl | and el over l S0 and over l S1 .

ISSN:-2230-7850
Fit the log-ratios log ( k ) and log (k )
(d) For k = 1,..., N , estimate w k =
a + b m , < 1
using log( k ) = 1 1 k k
a2 + b2 mk , k 1
r kk
1 + r kk
c
+
d
m
,
<1
1 k k
log( k ) = 1
c + d m , k 1
2 2 k
and
yk
Step 4. Apply the inverse wavelet transform.

V.
APPLICATION TO ULTRA SOUND IMAGES

We compare the performance of the proposed method to one conventional approach in speckle
filtering: the homomorphic Wiener filter. In particular, we apply Matlab's spatially adaptive Wiener filter
[8] to the image logarithm and subsequently perform the exponential transformation on the filtered out
put. The window size of the Wiener filter was experimentally optimized to produce the maximum output
SNR for each test image and for each amount of noise used in the simulations [6].
Fig 2, Fig 3 and Fig 4 shows the visual comparison for Ultrasound Chest, Gallbladder and Synthetic
images. Fig 5 and Fig 6 enables us to make a visual comparison of the results of the proposed method and
the homomorphic Wiener [2] filter on realistic noisy images. Fig 8 shows the horizontal line intensity
profile comparison for synthetic image.
Regarding the noise suppression performance, the proposed method shows a stable behavior with
respect to the tuning parameter K. Fig. 6 demonstrates that for different noise levels and for different test
images the same value of this parameter (K = 3) can be chosen to provide a nearly maximum output SNR.
It can also be seen that the window size 3X3 is optimal under the assumed speckle model.
10

ISSN:-2230-7850
Table 1 shows the quantitative comparison of widely used metrics, signal to noise ratio (SNR), and
peak signal to noise ratio (PSNR). The computation time for each algorithm is also included in this table.
We can notice that our proposed filter exhibits more than 2dB improvement in both SNR and PSNR over
Homomorphic Wiener filter [2]. The results clearly demonstrate that the proposed filter outperforms the
homomorphic spatially adaptive Wiener filtering both in terms of SNR and PSNR (Fig.10)
Fig.1 Gradual noise suppression in ultrasound Kidney image, using the proposed method
Fig. 2 Gradual noise suppression in ultrasound Chest image, using the proposed method
11

ISSN:-2230-7850
Fig.3 Visual Comparison for Ultrasound Chest image (a) original Image (b) speckled Noisy image (c)
Homomorphic spatially adaptive Wiener filter (d) proposed method, for K = 3 and window size 5x5
Fig.4 Visual Comparison for Ultrasound Gallbladder image (a) Original Image (b) artificially
speckled images, the results of the homomorphic spatially adaptive Wiener filter, and the results of
the proposed method, for K = 3 and window size 5x5
12

ISSN:-2230-7850
Fig.5 Results for synthetic Test image. (a) Synthetic test image (b) artificially speckled images, (c) the
results of the homomorphic spatially adaptive Wiener filter, and (d)the results of the proposed
method, for K = 7 and window size 3x3.
Fig. 6 Application to the real noisy image-1 (a) Real speckle noised Ultrasound image1 (b) Wiener filter
(c) proposed filter.
Fig. 7 Application to the real noisy image-2 (a) Real speckle noised Ultrasound image2 (b) Wiener filter
(c) proposed filter.
13

ISSN:-2230-7850
Fig.8.Horizontal profile comparison for synth image
Fig.9. The performance of the proposed method as a function of the value of the parameter K. (a) Test
image 1 from Fig.2, input SNR=13.6dB. (b) Test image 2 from Fig.2, input SNR=11.6 dB.
Fig.10. Comparison between the proposed filter and the homomorphic Wiener filter. (a) The synthetic
test image (b) The Test Image 1.
14

ISSN:-2230-7850
VI.
APPLICATION TO MAGNETIC RESONANCE IMAGES

Three clinical MR Images, a Pelvic MR Image of size 644X626, a Brain MR Image of size 471X341,
and a Spine MR Image of size 490X486 are used for the experimental evaluation purpose. In simulations,
complex zero mean white Gaussian noise with standard deviation c = 30 was added to these images.
Fig.12 to Fig14 shows the De-noising result of the proposed method comparison with the spatially
adaptive wiener filter. The results show that our proposed algorithm out performs with that of the
spatially adaptive Wiener filtering. Table.2 shows the quantitative comparison of widely used metrics,
signal to noise ratio (SNR), and peak signal to noise ratio (PSNR). The computation time for each algorithm
is also included in this table. We can observe that 5.35 dB improvement in both SNR and PSNR[6].
The quantitative performance of the method, for the 3x3 window size is illustrated in Fig.15. For
different noise levels the optimal value of Kis in the range [1.8, 2], and the algorithm shows a stable
behavior with respect to K. In our experiments, on a number of different reference MRI images, the
improvement over the spatially adaptive Wiener filtering was at least 0.5 dB (for relatively clean image)
and more than 3 dB for low SNR images. In Fig.15(b),we illustrate such a comparison between the two
filters for the reference Brain MR image. For the Wiener filter, the window size was optimized for each
noise level, to produce maximum output SNR. The application of the proposed method to real noisy MRI
images is demonstrated in Fig.11
The MRI images were provided by the Government General hospital, Vijayawada. The noise
suppression in these images is expected to facilitate further automatic processing, like window size is
illustrated in Fig.15. For different noise levels the optimal value of K is in the range [1.8, 2], and the
algorithm shows a stable behavior with respect to K. In our experiments, on a number of different reference
MRI images, the improvement over the spatially adaptive Wiener filtering was at least 0.5 dB (for
15

ISSN:-2230-7850
relatively clean image) and more than 3 dB for low SNR images. In Fig.15(b),we illustrate such a
comparison between the two filters for the reference Brain MR image. For the Wiener filter, the window
size was optimized for each noise level, to produce maximum output SNR. The application of the proposed
method to real noisy MRI images is demonstrated in Fig.11e.g.,segmentation.
Fig. 11. Application to real noisy MRI images. (a) and (c) original images, (b) and (d) the corresponding
results of the proposed method for K = 2 and a 3x3 window size.
Fig.15. Quantitative performance of the proposed method. (a) Influence of the parameter K for
different values c2 of the underlying complex Gaussian noise. (b) Noise suppression performance (for K
= 2) in comparison to the spatially adaptive Wiener filtering. The original MRI image from Fig.11 is used
in the simulations
16

ISSN:-2230-7850
Fig .12. Visual comparison for a Pelvic MR Image, Window 3X3, K=2 and = 30 (a) Original Image (b)
Noisy Image (c) Spatially Adaptive Wiener filter (d) Proposed filter.
Fig 13. Visual comparison for a Brain MR Image of size 471X341, Window 3X3, K=2and = 30 . (a)
Original Image (b) Noisy Image (c) Spatially Adaptive Wiener filter (d) Proposed filter.
17

ISSN:-2230-7850
Fig 14. Visual comparison for a Spine MR Image of size 490X486, Window 3X3, K=2and = 30 . (a)
Original Image (b) Noisy Image (c) Spatially Adaptive Wiener filter (d) Proposed filter.
VII. CONCLUSIONS
In this paper, we have developed a low-complexity De-noising algorithm using the joint statistics of
the wavelet coefficients and consider the dependencies between the coefficients. we have addressed the
development of new algorithms for improvement in image reconstruction by suppressing artifacts such as
blurring, noise, ring and other aliasing artifacts from sparse projections, commonly arise in CT, using
advanced wavelets such as quadratic spline wavelet transform. and demonstrated to be effectively
preserving the local features such as edges, corners and local orientations. These techniques are
important in medical diagnosis applications.
Further work focuses on the extension of multi resolution based denoising schemes to 3D-voxel data.
Additionally, a detailed comparison of the denoising schemes before and after a CT-reconstruction is
needed Conclusion.
18

ISSN:-2230-7850
Table .1:Signal to Noise Comparisons for the three Test Images.
Testimage1,5X5,K=3
Testimage2,5X5,K=3
Testimage3,5X5,K=7
Reconstruction
Method
SNR(dB)
PSNR
Comp.
SNR
PSNR
Comp.
SNR
PSNR(dB
(dB)
Time (Sec)
(dB)
(dB)
Time
(dB)
Comp.
Time(Sec)
(Sec)
Noisy image
17.75
25.06
---
17.71
27.31
---
23.89
32.60
Homomorphic
17.86
25.16
0.078
19.05
28.65
0.078
26.96
35.67
---
0.016
Wiener Filter
Proposed Filter
19.51
26.81
2.562
20.73
30.34
19
2.312
29.02
37.72
0.75

ISSN:-2230-7850
Table.2: Signal to Noise Ratio comparison for three images. Pelvic MR Image,
Brain MR Image and Spine MR Image, window size 3X3, K=2
Pelvic MR Image Size 644X626
Brain MR Image Size 471X341
Spline MR Image Size 490X486
=30
=30
=30
Reconstruction
Method
SNR(d
PSNR(dB
Comp.
SNR(d
PSNR(dB
Comp.
SNR(d
PSNR(dB
Comp.
B)
Time (Sec)
B)
Time
B)
Time
(Sec)
Noisy image
Spatially
adaptive
(Sec)
11.66
18.59
---
08.32
17.68
---
13.00
18.86
16.23
23.16
0.325
10.99
20.35
0.187
19.12
24.98
---
0.297
Wiener Filter
Proposed Filter
20.33
27.26
11.344
16.34
20
25.70
5.672
19.49
25.35
9.813

ISSN:-2230-7850
Wavelet used and
Improvement in
Computation Time
Noise type
SNR (dB)
(Sec)
Algorithm
Features
Quadratic spline
Coefficient classification technique. Adapts itself to
wavelet
Efficient
various types of noise and preference to expert.

Ultra Sound:
2.312
Applications: Speckle and Rician noise reduction.
Medical image
5.13
denoising
algorithm
Speckle noise and
Rician noise
MRI: 8.67
11.344
Table.3: Important features Versatile denoising algorithms for 512 X 512 Pelvic image with = 20 .
REFERNCES
[ 1] A. Pizurica, W. Philips, I. Lemahieu, and M. Acheroy, DespecklingSAR images using wavelets and a
new class of adaptiveshrinkage functions, Proc. IEEE Internat. Conf. on Image
Proc., ICIP 2001, Thessaloniki, Grece, Oct 2001.
[2] Achim A., Bezerianos A., and Tsakalides P., Novel Bayesian multiscale method for speckle removal in
medical ultrasound images, IEEE Trans. on Medical Imaging, vol. 20, no. 8, pp. 772-783, Aug 2001.
[3] A. Pizurica, W. Philips, I. Lemahieu, and M. Acheroy, A joint inter- and intrascale statistical model for
Bayesian wavelet based imagedenoising,IEEE Trans. Image Proc., vol. 11, no. 5, pp. 545557, May 2002.
[4]
BharathA.A.andJeffry.Ng,A Steerable Complex Wavelet Construction and its Application to Image
Denoising, IEEE Trans. on. Image Processing,Vol.14, No.7, July 2005.

21

ISSN:-2230-7850
[5]
Chang S.G., Yu B., and Vetterli M., Adaptive wavelet thresholding for image denoising and
compression, IEEE Trans. Image Proc., vol. 9, no. 9, pp. 1532-1546, Sept. 2000.
[6]Dr.M.Venugopal Rao, Advanced wavelet techniques for improved image reconstruction in Computer
Aided Tomography, osmania university, Phd Thesis,2011.
[7] X. Li and M. Orchard, Spatially adaptive denoising under overcomplete expansion, Proc. IEEE
Internat. Conf. on Image Proc., ICIP00, Vancouver, Canada, Sep 2000.
[ 8 ] Foucher S., Benie G. B., and Boucher J. M., Multiscale MAP filtering of SAR images,IEEE Trans.
Image Proc., vol. 10, no.10, pp. 4960, January 2001.
[9] Pizurica A., Image denoising using wavelets and spatial context modeling, PhD thesis, Ghent
University, Ghent, Belgium, 2002.
[10]
Bharath A.A. and Jeffry. Ng, A Steerable Complex Wavelet Construction and its Application to
Image Denoising, IEEE Trans. on. Image Processing, Vol.14, No.7, July 2005.
[11]
VenuGoapal Rao M., and Vathsal S., Wavelet methods for Noise reduction in Computed
Tomography, at Belarus-Indian Scientific and technical workshop, New Delhi, 24-26, Feb 2003.
[12]
VenuGoapal Rao M., and Vathsal S., Local Adaptive Bivariate Shrinkage Algorithm for Medical
Image Denoising, International Journal of Electronics Engineering (IJEE), 1(1), pp,59-65, Jan 2009.
22

Versatile Medical Image Denoising Algorithm

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Versatile Medical Image Denoising Algorithm

Enviado por

Direitos autorais:

Formatos disponíveis

Indian Streams Research Journal

Vol -3 , ISSUE 4, May.2013

Available online at www.isrj.net

AN EFFICIENT MEDICAL IMAGE DENOISING ALGORITHM

SRINIVASAKIRAN GOTTAPU AND M.VENUGOPAL RAO

Dept Of ECE , K.L University ,Andhra Pradesh , India.

Indian Streams Research Journal

Available online at www.isrj.net

Indian Streams Research Journal

Available online at www.isrj.net

elaborate the Summary of

Indian Streams Research Journal

Available online at www.isrj.net

BASIC THEORY OF PROPOSED ALGORITHM

Indian Streams Research Journal

Available online at www.isrj.net

Whereris the ratio of unconditional prior probabilities

Indian Streams Research Journal

Available online at www.isrj.net

THE EFFICIENT NOISE FILTERING TECHNIQUE

indicates the positions of significant wavelet coefficients (representing the

Indian Streams Research Journal

Available online at www.isrj.net

[8]. To initialize the classification, we

D = Y D , where J is the coarsest resolution level in the wavelet decomposition.

are computed from the histograms of {mk : k S0 } and {ek : k S0 }

Then the final estimation is defined as

Indian Streams Research Journal

Available online at www.isrj.net

In Fig. 1, we show an example of the empirical densities p M | X ( mk | xk ) and p E | X (ek | xk ) . The

} . We look for the polynomial coefficients a1, b1 , which minimize

Indian Streams Research Journal

Available online at www.isrj.net

And the polynomial coefficients a2 , b2 , which minimize

SUMMARY OF PROPOSED ALGORITHM

Step 1. Compute the non-decimated wavelet transform with J resolution levels.

Step 3. For each orientation D and for each scale

(a) For all the spatial positions k = 1, ..., N .

(ii) Compute the local spatial activity indicator

using averaged energy of the neighboring coefficients of

yk where the neighbors are the surrounding coefficients in a square window.

(b) Compute. r = lN=1 xl / N lN=1 xl

(c) Define S0 = {l : xl = 0} and S1 = {l : xl = 1} .

Indian Streams Research Journal

Available online at www.isrj.net

Fit the log-ratios log ( k ) and log (k )

(d) For k = 1,..., N , estimate w k =

Step 4. Apply the inverse wavelet transform.

APPLICATION TO ULTRA SOUND IMAGES

Indian Streams Research Journal

Available online at www.isrj.net

Indian Streams Research Journal

Available online at www.isrj.net

Indian Streams Research Journal

Available online at www.isrj.net

Indian Streams Research Journal

Available online at www.isrj.net

Fig.8.Horizontal profile comparison for synth image

Indian Streams Research Journal

Available online at www.isrj.net

APPLICATION TO MAGNETIC RESONANCE IMAGES

Indian Streams Research Journal

Available online at www.isrj.net