See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/343524224

Bio-Medical Image Denoising using Wavelet Transform

Article in International Journal of Recent Technology and Engineering (IJRTE) · May 2019

CITATIONS READS

11 522

2 authors:

Lalit Mohan Satapathy Pulen Das

Siksha O Anusandhan University Lakshmibai National University of Physical Education
30 PUBLICATIONS 57 CITATIONS 8 PUBLICATIONS 662 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Lalit Mohan Satapathy on 08 August 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1, May 2019

Bio-Medical Image Denoising using Wavelet

Transform
L. M. Satapathy, P. Das, A. Shatapathy, A. K. Patel

addition and communication errors. Thus, all medical

Abstract: Now a day’s images are used in various medical imaging devices need denoising technique to enhance the
science applications. A good quality image is highly required for
image quality which will help the doctors and medical experts
proper diagnosis. This paper proposes a novel hybridization
technique is proposed to improve the image quality of degraded for proper diagnosis. Traditional denoising techniques such as
image by the noise. The decomposition based proposed method Wiener filter, Median filter and wavelet soft threshold are
initially separates the image into four parts as LL, LH, HL and directly applied over noisy image for improving the quality of
HH. The high frequency information of the decomposed image is
de-noised using the conventional denoising techniques. The the image. The Wiener filter was proposed by Norbert
resultant image is reconstructed by applying the inverse wavelet Wiener which minimizes the mean square error by
transform. In this process the coefficients of the wavelet preserves computing the statistical estimate of an unknown signal using
the useful information corresponding to the image structure,
while suppresses the noisy elements. Experiments were conducted
random process. Due to wiener filter is dynamic and linear it
on various medical images available in public domain to compare provides point estimation, whereas it has a limitation to
the performance of the proposed algorithm with respect to the remove the multiplicative noise. The Median Filter is a
conventional methods such as Wiener filter, Median filter and
non-linear, edge preservative digital filtering technique
wavelet soft threshold. The validity of the presented approach is
subjectively quantified in terms of PSNR,MSE and structural widely used in the application field of image processing to
similarity. The experimental results demonstrate that the remove the noise. The main disadvantage in median filter is
proposed algorithm outperforms the existing denoising methods. its high computational complexity.In recent years, in the field
For medical images, the PSNR and the SSIM values improve by
using the proposed technique over various denoising approaches. of image denoising domain wavelet transform plays a vital
role. Wavelet transform are used as they have localization
Index Terms: DWT, IDWT, Image de noising, medical benefits in space domain and frequency domain. The packets
images, ssim . of wavelet allows flexible attainment for a given 2D signal.
Many methods are available for denoising of noisy images,
I. INTRODUCTION
but they have complexity in their methods as they are limited
Image denoising is a quality enhancement method in image to some extent. Traditional methods only de-noise the whole
processing, where the noise is removed from the noisy image image but it cannot de-noise the low frequency & high
and recovers the original image by retaining its quality, which frequency component separately due to which the images are
gets corrupted during its acquisition or transmission [1]. In changed abruptly. To overcome this problem, the high
medical field MRI, CT scan, ultrasound, x-ray etc. frequency component of the noisy image is filtered, and the
instruments are used for image acquisition in which noise can low frequency information is preserved. This motivated us to
be generated [2]. When noise is present in image degrades the apply discrete wavelet transform (DWT) based
objective quality and lowers down the clarity of low contrast decomposition to the image and as a result the low frequency
object [3]. Denoising is important in medical imaging as well as high frequency components are extracted and then
operations to recover the anatomical details that may be de-noise the high frequency component using traditional
suppressed in the data due to the noise [4].Bio-medical denoising filters. The remaining part of this paper is
images are normally - corrupted with noise; which degrade organized as follows: Section 2 provides a brief review of the
the useful detail of medical images which may affect the methodology. Section 3 demonstrated the detail experimental
diagnosis. As edges is most essential aspect for bio-medical results of the proposed algorithm, and the final conclusions
images, therefore the denoising needs to be balanced with are given in Section 4.
edge preservation. The main problem in bio-medical imaging
system is the adoption of the images obtained due to imperfect II. METHODOLOGY
The proposed algorithm gives a brief idea about
Revised Manuscript Received on May 28, 2019. decomposition-based image de-noising, where initially the
L. M. Satapathy, Department of Electrical and Electronics Engineering, noisy image(I) is decomposes into low frequency component
Siksha ‘O’ Anusandhan, Deemed to be University, Bhubaneswar, India as well as high-frequency components using discrete wavelet
P. Das , Department of EE, Indira Gandhi Institute of Technology,
Sarang, Odisha, India.
transform (DWT).
A Shatapathy, A K Patel, Department of Electrical and Electronics
Engineering, Siksha ‘O’ Anusandhan, Deemed to be University,
Bhubaneswar, India

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A9125058119/19©BEIESP 2479 & Sciences Publication
 Bio-Medical Image De Noising Using Wavelet Transform

The low frequency component contains the detail information analyses non-stationary signals. A set of mutually orthogonal
of the image whereas the high-frequency components of the wavelet basis functions are generated when the Discrete
digital image contains the approximation information [5]. The wavelet transform (DWT) decomposes a given signal. The
noise generally corrupts the high frequency components; as a discrete wavelet transform is invertible, so that the original
result, the image degrades and provides a wrong conclusion. image can be recovered from its discrete wavelet transform
The Fig.1 demonstrates the block representation of the DWT (DWT) representation.
based decomposition method of a noisy image. In Fig.2 the B. Noises.
results of the DWT based decomposition is shown and. Fig.3 Noise is an unacceptable signal that changes the property and
describes the spectral distribution of the low frequency and performance of the signal. Normally images are corrupted
high-frequency information of LL, LH, HL and HH. with noise likely Gaussian, Salt & pepper and speckle
distribution.
Original Image(O) 1) Gaussian noise
Gaussian noise is arising in electronic components which is
normally known as electronic noise. It is the demographical
∑ Noise(N) noise to that of the original distribution. The noise is
independent of each pixel as well as signal intensity and is
Noisy Image(I) preservative in nature [7].
The probability density function ‘g’ of a Gaussian random
variable ‘u’ is given by:
DWT
(𝑢−µ)2
1 −
g(u) = 𝑒 2𝜎2 (1)
𝜎√2𝜋
Where:
LL LH HL HH u: Grey level
Figure 1. Noisy Image decomposition using DWT. μ : Mean or average value
σ: Standard deviation

2) Salt & Pepper Noise

The other name of salt & pepper noise is impulsive noise[8].
It is generated during sharp and sudden disturbances in the
image and represents as white and black pixels. Median filter
Figure 2. (a) Original image (b) Decomposition of image Low-Low (c)
is mainly used for the noise removal of this type of noises.
Low-High (d) High-Low (e) High-High.
The probability density function ‘S’ of a Gaussian random
variable ‘u’ is given by:

𝑠𝑝 𝑓𝑜𝑟 𝑢 = 0 (𝑝𝑒𝑝𝑝𝑒𝑟)
S(u) = { 𝑠𝑠 𝑓𝑜𝑟 𝑢 = 2𝑛 − 1(𝑠𝑎𝑙𝑡) (2)
1 − (𝑠𝑝 + 𝑠𝑠 ) 𝑓𝑜𝑟 𝑢 = 𝑘(0 < 𝑘 < 2𝑛 − 1)

Figure 3. Spectrum of (a) Low-Low (b) Low-High (c) High-Low (d) 3) Speckle Noise
High-High. Another type of noise which corrupts the quality of the active
radar, medical images and optical coherence tomography
To de-noise the noisy image the conventional denoising images is known as speckle noise [9].
technique is applied to the HH. When the HH is processed The major cause of speckle noise generation is the effect of
through the conventional filter HH1 is produced. The new or environmental conditions during imaging sensor in the
de-noised image is generated by applying Inverse discrete process of image transmission.
wavelet transform (IDWT) to LL, LH, HL and HH1[6]. The Speckle noise follows gamma distribution which is shown
proposed algorithm highlights that conventional filters when below:
applied with DWT performs better as compare to the direct 𝑢
𝑢𝛼−1 −
application of the conventional technique. F(u) =
(𝛼−1)!𝑎𝛼
𝑒 𝑎 (3)
A. Discrete wavelet transform
The discrete wavelet transforms (DWT) [5] is an effective and Where:
useful tool for decomposing the 2D signal. It provides the u: Grey level
time and frequency domain representation. The development 𝑎𝛼 : Variance
of discrete wavelet (DWT) was to overcome the short coming
of the Short time Fourier transform (STFT), the use of that to

Published By:
Retrieval Number: A9125058119/19©BEIESP Blue Eyes Intelligence Engineering
2480 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1, May 2019

C. Filters. III. EXPERIMENTAL RESULTS

In the recent years the big challenge for the researcher to
3.1 Performance evaluation measure
de-noise the image to recover the existing image which was
The proposed algorithm is verified using subjective analysis
corrupted or blurred. It is necessary step used before the
approach. The performance evaluator such as SSIM, PSNR
image detail is analyzed. It is important to use a better
and MSE are considered to quantify the result.
denoising method to recover the image from the data
corruption.
1) Median Filter
Median filter is a nonlinear filter which is used for denoising
the image [10]. It works efficiently for removal of salt &
pepper noise. The objective of filtering in digital image is
denoising as well as edge preservation.
2)Wiener Filter
A Wiener filter which is a flexible low-pass filter uses pixel
wise adoption. This is an adaptive filtering method with linear
in nature [11]. The method used in wiener filter is a statistics
approximation-based approach from nearer of each pixel. The
main advantage of this filter is that it preserves the edges of an
image. Figure 4. Performance of MRI brain images (a) Input image (b) Noisy
image (Gaussian noise with variance 0.01) (c) applying median filter
The Wiener filter is: (d) Applying wiener filter (e) Applying soft threshold (f) Proposed
k∗(a,b) 𝑃𝑠 (a,b) method.
w (a, b) = (4)
|k (a,b)|2 𝑃𝑠 (a,b)+𝑃𝑛 (a,b)
Dividing through by 𝑃𝑠 makes its behaviour easier to explain:

k∗(a,b)
w (a, b) = 𝑃 (a,b) (5)
|k (a,b)|2 + 𝑛
𝑃𝑠 (a,b)

where,
k (a, b) = Degradation function
k*(a, b) = Complex conjugate of degradation function
𝑃𝑛 (a, b) = Power Spectral Density of Noise
𝑃𝑠 (a, b) = Power Spectral Density of un-degraded
image

3) Wavelet Soft Thresholding Figure5. Performance of MRI brain images (a) Input image (b) Noisy
image (Salt & Pepper noise) (c) applying median filter (d) Applying
Soft thresholding is a delete and scaling method where the wiener filter (e) Applying soft threshold (f) Proposed method.
coefficients less than the threshold are deleted and scales that
which are left[12]. It is also known as wavelet shrinkage. In
comparison to hard thresholding which keeps or removes
values of coefficients. The formula for the universal threshold
is expressed as follows:

λ = σ√2ln(𝑀) (6)

Where,
σ: Average variance of the noise
M: Signal length
σ is calculated using median estimate method as follows:

𝑀𝑒𝑑𝑖𝑎𝑛(|𝑊1,𝐾 |)
σ= (7) Figure 6. Performance of MRI brain images (a) Input image (b) Noisy
0.6745
Soft thresholding function is defined as: image (Speckle noise) (c) applying median filter (d) Applying wiener
𝑠𝑔𝑛(𝑊𝑗,𝑘) (|𝑊𝑗,𝑘 | − λ); |𝑊𝑗,𝑘 | ≥ λ filter (e) Applying soft threshold (f) Proposed method.
𝑊̂𝑗,𝑘 = {
0; |𝑊𝑗,𝑘 | < λ

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A9125058119/19©BEIESP 2481 & Sciences Publication
 Bio-Medical Image De Noising Using Wavelet Transform

= 20 log10(MAX) - 10 log10(MSE) (13)

Where,
MAX: Maximum possible pixel value of the image.
MSE: Mean Square Error [13].

c)Mean square error(MSE)

The Mean Square Error(MSE) [13] is the cumulative error
between the compressed image and the original image. The
lower the MSE, better the quality of the compressed image. It
Figure 7. Histogram of MRI brain images (a) Input image (b) Noisy is calculated as:
image (Gaussian noise with variance 0.01) (c) applying median filter (d) 1
Applying wiener filter (e) Applying soft threshold (f) Proposed method. MSE = ∑𝑝−1 𝑞−1
𝑎=0 ∑𝑏=0[𝐼(𝑎, 𝑏) − 𝐾(𝑎, 𝑏)]
2
(14)
𝑝𝑞

a) Structural Similarity Index (SSIM): Where p q: Dimension of the image.

Structural Similarity Index (SSIM) is a perceptual metric that I (a, b): Intensity of pixels (a, b) in original image.
quantifies image quality degradation which is caused by K (a, b): Intensity of pixels (a, b) in de-noised image.
processing such as data compression or by losses in data
transmission. It is used for measuring the similarities between 3.2. Experimental Results
two images. It is a full reference metric i.e. the measurement To validate the proposed algorithm the experiment was
of quality of image is based on an initial uncompressed or conducted and recorded over 30different images. These
distortion free image as reference. SSIM quality assessment images are selected from the open source. To quantify the
index is based on the computation of three terms, namely the result average performance of subjective parameters such as
luminance term, the contrast term and the structural term. The PSNR, MSE and SSIM are calculated. The results were
overall index is a multiplicative combination of the three compared with state of art traditional de-noising methods.
terms[14]. Fig. 4, 5 and 6 demonstrates the behaviors of different noises
SSIM (a, b) = [q (a, b)] α. [w (a, b)] β. [e (a, b)] γ as Gaussian noise, Salt & Pepper noise, Speckle noise on an
2µ𝑎 µ𝑏 +𝐶1 image. Then the traditional techniques as (a) median filter (b)
Where q (a, b) = (8) wiener filter and (c) soft threshold applied and results are
µ𝑎 2 +µ𝑏 2 +𝐶1
incorporated for visual analysis purpose. Results of proposed
2𝛿𝑎 𝛿𝑏 +𝐶2 algorithm are shown in fig.4-6 and for better comparison
w (a, b) = (9)
𝛿𝑎 2 +𝛿𝑏 2 +𝐶2 Fig.7. is added where histogram distribution of all the
methods as well as original image are demonstrated.
𝛿𝑎𝑏 +𝐶3
e (a, b) = (10) The proposed method gives better distribution and avoids the
𝛿𝑎 𝛿𝑏 +𝐶3
Where µ𝑎 : Local mean for image a. stretching problems. It also preserves the brightness of the
image.
µ𝑏 : Local mean for image b.
Higher value of PSNR in the proposed method concludes that
𝛿𝑎 : Standard deviation for image a.
the conventional method with DWT performs better noise
𝛿𝑏 : Standard deviation for image b.
reduction [Fig.8]. The mean square error of Fig.9 shows that
𝛿𝑎𝑏 : cross-covariance for images a, b.
If α = β = γ = 1 and C3 = C2/2 then the above index is the proposed method has less error as compare to the other
simplifying to: methods. The structural similarity index demonstrates the
quality of the de-noised image. Fig.10 indicates that the SSIM
(2µ𝑎 µ𝑏 +𝐶1 )(2𝛿𝑎 𝛿𝑏 +𝐶2 )
SSIM(a,b)= (11) of the DWT based noise reduction methods are higher as
(µ𝑎 2 +µ𝑏 2 +𝐶1 )(𝛿𝑎 2 +𝛿𝑏 2 +𝐶2 )
compare to the conventional methods.
b)Peak signal to noise ratio(PSNR) The experiments are also conducted over different images
PSNR [13] is the ratio between the maximum possible power corrupted by Salt & Pepper noise and Speckle noise to
of a signal and the power of corrupting noise. PSNR measures validate the proposed algorithm. It was found that the method
the peak signal to noise ratio between two images which is outperforms on the case of the above-mentioned noise and the
used as quality measurement between two images (i.e. evidence is demonstrated in fig.11-13. In the case of PSNR it
original image and compressed image). Higher the value of was evidence that the wiener filter with DWT performs better
PSNR, better the quality of the compressed image. PSNR is in the case of Salt & Pepper noise and Speckle noise where as
usually expressed in terms of the logarithmic decibel scale. Median filter with
PSNR is calculated as:
𝑀𝐴𝑋 2
PSNRdB = 10 log10( ) (12)
𝑀𝑆𝐸

Published By:
Retrieval Number: A9125058119/19©BEIESP Blue Eyes Intelligence Engineering
2482 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1, May 2019

DWT gives better result in the case of Speckle noise. In the

case of MSE it was found that the proposed method has low
values as compared to the traditional methods [Fig.12].

Figure 12. MSE comparison of proposed algorithm & traditional

method using various denoising schemes for different type of noises.
Figure 8. PSNR comparison of variances using different denoising
schemes for Gaussian noise.

Figure 9. MSE comparison of variances using different denoising Figure 13. SSIM comparison of proposed algorithm & traditional
schemes for Gaussian noise. method using various denoising schemes for different type of noises.

The SSIM of the proposed method evidence that the

denoising with DWT method has high similarity value to the
original image.

IV. CONCLUSION
The wavelet-based hybrid denoising method is introduced in
this paper. This method is implemented and tested on number
Figure 10. SSIM comparison of variances using different denoising
of images available publicly. The main objective behind the
schemes for Gaussian noise. proposed method is to decompose the image before
denoising.

Only the high frequency information is de-noised using

traditional methods and the low frequency information are
preserved to recovery the original image. In this method, the
image quality is improved in terms of detail and edge, and the
noise was removed. Using traditional denoising methods with
the decomposition technique gives better result as compare to
the traditional methods only. The proposed algorithm
concludes that the method has better denoising effect on
medical images.

REFERENCES
1. D. L. DONOHO, "DE-NOISING BY SOFT-THRESHOLDING," IN IEEE
TRANSACTIONS ON INFORMATION THEORY, VOL. 41, NO. 3, PP. 613-627,
MAY 1995.
Figure 11. PSNR comparison of proposed algorithm & traditional
method using various denoising schemes for different type of noises.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A9125058119/19©BEIESP 2483 & Sciences Publication
 Bio-Medical Image De Noising Using Wavelet Transform

2. R. D. Nowak, "Wavelet-based Rician noise removal for magnetic

resonance imaging," in IEEE Transactions on Image Processing, vol.
8, no. 10, pp. 1408-1419, Oct. 1999.
3. R. Gonzalez and R. Woods, “Digital Image Processing using
MATLAB,” Second Edition, THM (2009).
4. S. Kara and F. Dirgenali, “A system to diagnose atherosclerosis via
wavelet transforms, principal component analysis and artificial neural
networks,” Expert Systems with Applications, vol. 32, no.2, pp.
632-640, 2007.
5. S. G. Mallat and S. Zhong, “Characterization of signals from multiscale
edges,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 14, no. 7, pp.
710– 732, Jul. 1992.
6. Sung, T. "Low-power and high-performance 2-D DWT and IDWT
architectures based on 4-tap Daubechies filters." WSEAS
TRANSACTIONS ON ELECTRONICS, vol. 4, no. 1. pp. 50-55,2007.
7. Selami, A.M.A.A. and Fadhil, A.F.,“A Study of the Effects of Gaussian
Noise on Image Features”. Kirkuk university journal for scientific
studies, vol.11, no.3, pp.152-169, 2016
8. Liu Ying-hui, Gao Kun and Ni Guo-qiang, "An improved trilateral filter
for Gaussian and impulse noise removal," 2010 The 2nd International
Conference on Industrial Mechatronics and Automation, Wuhan,
2010, pp. 385-388.
9. H. Rabbani, M. Vafadust, P. Abolmaesumi and S. Gazor, "Speckle
Noise Reduction of Medical Ultrasound Images in Complex Wavelet
Domain Using Mixture Priors," in IEEE Transactions on Biomedical
Engineering, vol. 55, no. 9, pp. 2152-2160, Sept. 2008.
10. HakanGüraySenel, Richard Alan Peters and Benoit Dawant. 2002.
Topological Median Filter. IEEE Trans on Image Processing. 11(2):
89-104.
11. H. Yoshino, C. Dong, Y. Washizawa and Y. Yamashita. “Kernel
Wiener Filter and its Application to Pattern Recognition”, IEEE
Transactions on neural networks, 2010.
12. Byung-Jun Yoon and P. P. Vaidyanathan, "Wavelet-based denoising
by customized thresholding," 2004 IEEE International Conference on
Acoustics, Speech, and Signal Processing, Montreal, Que., 2004, pp.
ii-925.
13. L. M. Satapathy, A. Dalai, S. Satapathy and A. Jena, "Satellite image
enhancement based on multi-technology fusion," 2018 Second
International Conference on Inventive Communication and
Computational Technologies (ICICCT), Coimbatore, 2018, pp.
1677-1680.
14. L.M. Satapathy, R.K. Tripathy, & P. Das, “A Combination of
Variational Mode Decomposition and Histogram Equalization for
Image Enhancement”Natl. Acad. Sci. Lett. (2018).
https://doi.org/10.1007/s40009-018-0742-y

Published By:
Retrieval Number: A9125058119/19©BEIESP Blue Eyes Intelligence Engineering
2484 & Sciences Publication
View publication stats