International Journal of Integrated Engineering, Vol. 6 No. 3 (2014) p.
51-56
BL_Wiener Denoising Method for Removal of Speckle Noise
in Ultrasound Image
Suhaila Sari1,2,3,*, Zuliana Azreen Zulkifeli1, Hazli Roslan4, Nabilah Ibrahim1,2
1
Faculty of Electrical and Electronic Engineering,Universiti Tun Hussein Onn Malaysia, 86400 Johor, MALAYSIA.
2
Biomedical Engineering Modelling and Simulation Research Group (BioMems), Universiti Tun Hussein Onn
Malaysia, 86400 Johor, MALAYSIA.
3
Artifical Intelligence and Computing Vision Research Group (AICoV), Universiti Tun Hussein Onn Malaysia, 86400
Johor, MALAYSIA.
4
Faculty of Engineering Technology, Universiti Tun Hussein Onn Malaysia, 86400 Johor, MALAYSIA.
Abstract: Medical imaging techniques are extremely important tools in medical diagnosis. One of these important
imaging techniques is ultrasound imaging. However, during ultrasound image acquisition process, the quality of
image can be degraded due to corruption by speckle noise. The enhancement of ultrasound images quality from
the 2D ultrasound imaging machines is expected to provide medical practitioners more reliable medical images in
their patients’ diagnosis. However, developing a denoising technique which could remove noise effectively without
eliminating the image’s edges and details is still an ongoing issue. The objective of this paper is to develop a new
method that is capable to remove speckle noise from the ultrasound image effectively. Therefore, in this paper we
proposed the utilization of Bilateral Filter and Adaptive Wiener Filter (BL_Wiener denoising method) for images
corrupted by speckle noise. Bilateral Filter is a non-linear filter effective in removing noise, while Adaptive
Wiener Filter balances the tradeoff between inverse filtering and noise smoothing by removing additive noise
while inverting blurring. From our simulation results, it is found that the BL_Wiener method has improved
between 0.89 [dB] to 3.35 [dB] in terms of PSNR for test images in different noise levels in comparison to
conventional methods.
Keywords: Ultrasound imaging, denosing technique, speckle noise, Bilateral Filter, Adaptive Wiener Filter;
BL_Wiener
1. Introduction scatterers reflect the sound waves towards the sensor. The
degradation effects including suppression of edges, and
Medical imaging techniques are extremely important
structural details, which may affect image-based
tools in medical diagnosis. One of these important
diagnosis decision. Therefore, noise reduction for medical
imaging techniques is ultrasound imaging. Ultrasound is a
images is important to improve the image. Even though
form of non-ionizing radiation and considered safe to
there are sophisticated technologies such as 3D and 4D
users. Ultrasound imaging is highly non-invasive where it
ultrasound imaging machine, MRI and CT scan
will not entering or penetrating into body or disturbing
machines, the availability of these machines are limited to
body tissue especially in a diagnosis [1]. Ultrasound
certain medical centers. These sophisticated machines are
imaging is based on the transmission towards human
mostly available only at public medical centers or
body of longitudinal acoustic waves with a frequency and
university’s hospitals in capital urban areas and private
the reception of produced echo [2]. The signal is created
medical centers. Furthermore, the cost for these medical
and then detected by piezo-electric effect. Ultrasound
imaging is significantly higher in comparison to 2D
imaging has several advantages over some other medical
ultrasound imaging. The enhancement of ultrasound
imaging techniques in terms of cost, safety and allows
images quality from the 2D ultrasound imaging machines
real time imaging. However, during ultrasound image
is expected to provide medical practitioners more reliable
acquisition process, the quality of image can be degraded
medical images in their patients’ diagnosis. Through a
due to corruption by noise. Noise may be generated due
more accessible medical facility, it will benefit the public
to imperfect instruments used in image processing,
for earlier and faster preventions and treatments.
problems with data acquisition process and interferences
Many denoising methods have been proposed over
and as well as the influence of the anatomy of the tissues
the years, such as the Wiener Filter, Wavelet
in the body as fat. In ultrasound imaging, images are
Thresholding [3] [4], anisotropic filtering [5], Total
usually corrupted by speckle noise, which is a type of
Variation method [6], and Non-Local Means methods [7].
multiplicative noise. Speckle noise is generated due to
scattering phenomenon. A number of elementary
*Corresponding author: suhailas@[Link]
2015 UTHM Publisher. All right reserved. 51
[Link]/ojs/[Link]/ijie
S. Suhaila et al., Int. J. Of Integrated Engineering Vol. 6 No. 3 (2014) p. 51-56
However, developing a denoising technique which could domain and second filter works in intensity domain. This
remove noise effectively without eliminating the image’s filter applies spatially weighted averaging smoothing
edges and details is still an ongoing issue in image edges. It replaces the pixel value at corresponding pixel
processing. In the Wavelet Threshold, a signal is with an average of similar and nearby pixel values. The
decomposed into approximation and detail subbands, and BF is defined as:
the coefficients in the detail subbands are processed via
hard or soft thresholding [8]. The main task of the (1)
wavelet thresholding is the selection of threshold value.
The Bilateral Filter was proposed by Tomasi et al [9]. It
applies spatially weighted averaging. Bilateral Filter is where
capable to denoise noise from a corrupted image while I filtered is the filtered image
preserving the edge of the image through the non-linear I is the original input image to be filtered
combination of adjacent images. This technique is non- x is the coordinates of the current pixel to be filtered
iterative, local and easily produced. Bilateral filter Ω is the window centered in x
provides good filtering performance for grayscale images fr is the range kernel for smoothing differences in
and color images. Referring to Zhang [10], Adaptive intensities. This function can be a Gaussian function
Bilateral Filter (ABF) is proposed. According to G.R. gs is the spatial kernel for smoothing differences in
Sinha [11] in 2012, the bilateral filter has been applied to coordinates. This function can be a Gaussian function
denoise medical image.
Combination of Bilateral Filter and other filter to Several characteristics of BF which explains its
denoise image has been introduced in many papers. success are (i) BF is simple to formulate it. Each pixel is
Bilateral Filter and Wavelet Thresholding was applied by replaced by a weighted average of its neighbors, (ii) BF
Zhang and Guntruk to eliminate noise in real noisy depends only on two parameters that indicate the size and
images [12]. This method is also used by Sudipta Roy, contrast of the features to preserve, (iii) BF is a non-
NidulSinha and Asoke K. Sen to eliminate noise in iterative technique. It is easy to set so that the parameters
medical image [13]. Through these researches, the effect is not cumulative over several iterations. Although
combination of BF with other denoising technique has BF is effective in removing noise, however the drawback
proven significant improvement in noise reduction. of this filter is that it may oversmoothes the edges and
Therefore, in this paper we proposed the utilization of BF details at the same time.
and Wiener filter for images corrupted by speckle noise. Due to this shortcome, Wiener Filter (WF) [15] is
proposed to be used to further denoising while preserving
The Wiener filter is an established filtering method
remaining edges and details. The WF is first introduced
and is best known for linear image denoising [14 ]. It
for Gaussian noise removal. It plays an important role in
balances the tradeoff between inverse filtering and noise
a variety of applications such as linear prediction, image
smoothing by removing additive noise while inverting
processing, echo cancellation, signal recovery, channel
blurring. The Wiener filter was proposed by Nobert
equalization and system identification. WF coefficients
Wiener [15]; since then, it has been improved and
are calculated to minimize the average squared distance
implemented in various domains such as the spatial [16],
between the filter output and the desired signal. Basically,
frequency [17], [18] domains.
Wiener theory assumes that the signal is stationary
The objective of this paper is to develop a new
process. However, if the filter coefficients are
method that is capable to remove speckle noise from the
periodically recalculated for each block of signal samples
ultrasound image effectively. This proposed method
N then filter adapt to the characteristics of the average
utilizes the advantages of both Bilateral Filter and Wiener
signal in the block and the block-adaptive, which is
Filter. Therefore the proposed method is expected to
namely Adaptive Wiener Filter (AWF) [19]. This low
contribute effective preservation of fine details and edges
pass filter is applied in a local neighborhood of 3 × 3
in the image while removing the noise. The proposed
pixels blocks of the image. The AWF estimated the local
method is denoted as BL_Wiener method, hereafter. The
mean and variance around each corresponding pixel. The
denoising is performed to 2D ultrasound images
AWF is defined as:
corrupted by speckle noise by using MATLAB software.
The paper is organized as follows. The proposed
BL_Wiener method is described in Section 2. The results S(i,j) = W (i,j) * X(i,j) (2)
and analysis are discussed in Section 3. Finally,
concluding remarks are drawn in Section 4. where
S (i, j): Images that have been filtered
W (i, j): Wiener filter
2. Proposed BL_Wiener Denoising Method X (i, j): Image corrupted by noise
Bilateral filter (BF) [9] has been proposed for gray
scale images and color images. BF is an effective filter in The utilization of both techniques is expected to
smoothing noise especially in smooth regions in a provide a better denoising performance which is capable
corrupted image. BF is achieved by the combinations of to remove noise effectively while preserving edges and
spatial Gaussian filters. One of filter works in spatial details.
52
S. Suhaila et al., Int. J. Of Integrated Engineering Vol. 6 No. 3 (2014) p. 51-56
Noisy 2D ultrasound image
Bilateral Filter
Adaptive Wiener Filter
Restored 2D ultrasound image
(a) Original image (5-6 week pregnancy)
Fig. 1 Block diagram of BL_Wiener denoising method.
The BL_Wiener technique is tested to ultrasound
images. The analysis for restored images is performed by
objective and subjective evaluation methods. The
subjective evaluation method is performed by evaluating
the visual effects. The objective evaluation method is
using the Peak Signal to Noise Ratio (PSNR) [11]. The
calculation of PSNR is calculated by using two images;
of the original image (clean) and the denoised image.
Higher PSNR value means higher image quality
produced, conversely lower MSE means better quality
restored image. The PSNR is defined as:
(b) Image (a) corrupted by low speckle noise (2e2)
(255 X 255) (3)
PSNR [dB] = 10 log10
MSE
MSE =
∑ (predicted value - real value) 2
(4)
resolution x × resolution y
where
MSE: mean squared error
3. Results and Discussion
In this paper, the performance of the BL_Wiener is
compared with conventional denoising methods, which
are BF and AWF by using MATLAB software. The (c) Image (a) corrupted by medium speckle noise (7e11)
parameter settings for BF are Ω = 10, g = 2, f = 0.2,
respectively. On the other hand, the adaptive filter size
for the AWF is set to 3 x3. The BL_Wiener utilized the
similar parameter settings as the BF and AWF. For
simulation, original 2D ultrasound images (i.e. image
captured without corrupting with additional noise) are
corrupted with speckle noise in different levels: low
(2e2), medium (7e11) and high (8e20) noise.
In this simulation, several 2D ultrasound images
have been used to represent different type ultrasound
images. The examples of the test images are 5-6 week
pregnancy image and kidney image.
Figure 2 shows the image of 5-6 pregnancy that has
(d) Image (a) corrupted by high speckle noise (8e20)
been corrupted with different level of speckle noise, used
as test images to investigate the effectiveness of proposed
Fig. 2 (a) Original image (i.e. image captured without
technique. The objective evaluation results are
corrupting with additional noise) and corrupted with
represented in Tables 1 and 2.
speckle noise in different levels.
53
S. Suhaila et al., Int. J. Of Integrated Engineering Vol. 6 No. 3 (2014) p. 51-56
Table 1. Comparison of denoising performance for 5-6
week pregnancy 2D ultrasound image in terms of PSNR.
Difference
Types of Speckle
PSNR Ave- (Proposed- (i)
denoising noise
[dB] rage Compared
method level
method)[dB]
(ii)
2e2 75.62
BF 7e11 75.59 75.60 1.09
8e20 75.58
(a)
2e2 73.35
AWF 7e11 73.31 73.33 3.35
8e20 73.34
2e2 76.66
BL_
7e11 76.69 76.69 0.00
Wiener
8e20 76.71
From Table 1, it is clear that the BL_Wiener
provided better performance to remove the speckle noise.
(b)
BL_Wiener produced higher PSNR in comparison with
conventional methods (Average of 3.35 [dB] better than
AWF and 1.09 [dB] higher than BF, respectively).
Table 2. Comparison of denoising performance for
kidney 2D ultrasound image in terms of PSNR.
Difference
Types of Speckle
PSNR Ave- (Proposed-
denoising noise
[dB] rage Compared
method level
method)[dB]
2e2 70.33 (c)
BF 7e11 70.31 70.31 0.89
8e20 70.29
2e2 69.88
AWF 7e11 69.88 69.87 1.32
8e20 69.85
2e2 71.20
BL_
7e11 71.20 71.19 0
Wiener
8e20 71.18
(d)
From Table 2, it is clear that the BL_Wiener Fig. 3 (a) Original image (i.e. image captured without
provided better performance to remove the speckle noise. corrupting with additional noise) for 5-6 week pregnancy
BL_Wiener produced higher PSNR in comparison with 2D ultrasound image, (b) denoised image using AWF, (c)
conventional methods (Average of 1.32 [dB] improved denoised image using BF, (d) denoised image using
than AWF and 0.89 [dB] higher than BF, respectively). BL_Wiener technique.
54
S. Suhaila et al., Int. J. Of Integrated Engineering Vol. 6 No. 3 (2014) p. 51-56
(a) (b)
(a)
(c) (d)
Fig. 4 (a) Enlarged smooth region in Fig. 3(a), (b)
enlarged smooth region in Fig. 3(d), (c) enlarged edges
region in Fig. 3(a), (d) enlarged edges region in Fig. 3(d).
(b)
Fig. 3 and 4 show the image of 5-6 pregnancy (image
captured without corrupting with additional noise) that
has been used as a test image in this study. This type of
image is chosen because to make sure that this method
can be applied to trimester pregnancy image diagnosis.
From visual observation, overall image noise has been
smoothed compared in result of BL_Wiener (Fig. 3(d)
and Fig. 4(b)) but at same time strong edges and details
are still preserved (Fig. 3(d) and Fig. 4(d)).
Figure 5 shows the image of kidney (image captured
without corrupting with additional noise) that has been
used as a test image in this study. This types of image is
chosen to investigate the whether this method can be
applied to detect the location of tumor for diagnosis. (c)
From visual inspection, we can observe that noise is
reduced but at the same time it still preserves the strong
edges and details of the image.
From our investigation, we found that by applying
both Bilateral Filter and Adaptive Wiener Filter to a 2D
ultrasound image, it has produced a less noisy but edges
and detail preserved output image. Moreover, our
BL_Wiener has higher performance in terms of PSNR
evaluation in comparison to conventional denoising
methods.
(d)
Fig. 5 (a) Original image (i.e. image captured without
corrupting with additional noise) for kidney 2D
ultrasound image, (b) denoised image using AWF, (c)
denoised image using BF, (d) denoised image using
BL_Wiener technique.
55
S. Suhaila et al., Int. J. Of Integrated Engineering Vol. 6 No. 3 (2014) p. 51-56
4. Summary [9] Tomasi, C., and Manduchi, R. Bilateral filtering
for gray and color images. Proceeding of
In conclusion, the BL_Wiener is proposed to denoise
International Conference in Computer Vision,
speckle noise existing in 2D ultrasound images while
(1998), pp. 839-846.
preserving the original image edges and details by the
utilization of Bilateral Filter and Adaptive Wiener Filter.
[10] Zhang, B. Adaptive bilateral filter for sharpness
From our investigation, it is found that the BL_Wiener
enhancement and noise removal. IEEE
has improved between 0.89 [dB] to 3.35 [dB] in terms of
Transaction on Image Processing, Volume 17,
PSNR for test images in different noise levels in
(2008), pp. 664-678.
comparison to conventional methods.
[11] Bhonsle, D., Chandra, V., and Sinha, G.R. Medical
Acknowledgements image denoising using bilateral filter. International
The authors would like to thank Universiti Tun Journal of Image, Graphic and Signal Processing,
Hussein Onn Malaysia (UTHM) and Malaysian Volume 6, (2012), pp. 36-43.
Government for the support and of this study.
[12] Zhang, M., and Gunturk, B. A new image
denoising method based on the bilateral filter.
References Procceeding of IEEE International Conference of
Acoustic, Speech and Signal Processing, (2008),
[1] U. S. Department of Health & Human Services. pp. 929-932.
Ultrasound imaging. [Link]
Emitting Product and Procedure/ Medical Imaging/ [13] Roy, R., Sinha, N., and Asoke, K.S. An efficient
[Link]. denoising model based on wavelet and bilateral
filters. International Journal of Computer
[2] Narayanan, S.K., and Wahidabanu, R.S.D. A view Applications, Volume 53(10), (2012).
on despeckling in ultrasound imaging.
International Journal of Signal Processing Image [14] Gonzalez, R.C., Woods, R.E., and Eddins, S.L.
Processing and Pattern Recognition, Volume 2, Digital image processing using MATLAB.
(2009), pp. 85-98. Prentice Hall, (2004).
[3] Donoho, D.L., and Johnstone, I.M. Ideal spatial [15] Wiener, N. Extrapolation, interpolation, and
adaptation by wavelet shrinkage. Biometrika, smoothing of stationary time series. Wiley, (1949).
Volume 81(3), (1994), pp. 425-455.
[16] Jin, F., Fieguth, P., Winger, L., and Jernigan, E.
[4] Ogawa, K., Sakata, M., and Li, Y. Adaptive noise Adaptive Wiener filtering of noisy images and
reduction of scintigrams with a wavelet transform. image sequences. Proceeding of IEEE
International Journal of Biomedical Imaging, International Conference in Image Processing,
Volume 2012, (2012), http:// [Link]/ 10.1155/ (2003), pp. 349-352.
2012/ 130482.
[17] Suhaila, S., and Shimamura, T. Image restoration
[5] Perona, P., and Malik, J. Scale-space and edge based on edgemap and Wiener filter for preserving
detection using anisotropic diffusion. IEEE fine details and edges. International Journal of
Transactions of Pattern Analysis and Machine Circuits, Systems and Signal Processing, Volume
Intelligence, Volume 12(7), (1990), pp. 629–639. 6(5), (2011), pp. 618-626.
[6] Buades, A., Coll, B., and Morel, J. Neighborhood [18] Ercelebi, E., and Koc, S. Lifting-based wavelet
filters and PDE’s. Numerische Mathematik, domain adaptive Wiener filter for image
Volume 105(1), (2006), pp. 1-34. enhancement. Proceeding of IEEE Vision, Image
and Signal Processing, Volume 153(1), (2006), pp.
[7] Rudin, L. I., Osher, S., and Fatemi, E. Nonlinear 31-36.
total variation based noise removal algorithms.
Physica D, Volume 60(1-4), (1992), pp. 259-268. [19] Lim, J.S. Two Dimensional Signal and Image
Processing. Prentice Hall, (1990).
[8] David, D.L., and Johnstone, I.M. Adapting to
unknown smoothness via wavelet shrinkage. [20] Tewari, K., and Tiwari, M.V. Efficient removal of
Journal of American Statistical Association, noise in digital image. International Journal of
Volume 90(432), (1995), pp. 1200-1224. Scientific and Research Publications, Volume 2,
(2012), pp. 1-7.
56