You are on page 1of 4

Speckle Noise Reduction in Ultrasound Images

using SRAD and Guided Filter

Hyunho Choi Jechang Jeong


Department of Electronics and Computer Engineering Department of Electronics and Computer Engineering
Hanyang University Hanyang University
Seoul, Korea Seoul, Korea
hyunho0619@gmail.com jjeong@hanyang.ac.kr

Abstract— Ultrasound imaging devices are more secure and filtering methods have been proposed. These techniques such
real-time than other medical diagnosis devices such as X-ray, CT, as non-local means (NLM) filter [3], anisotropic diffusion (AD)
and MRI. However, due to speckle noise, it exhibits lower filter [4], and bilateral filter [5] have been proposed, and have
resolution and image quality than other than medical diagnostic been developed to eliminate speckle noise while minimizing
devices. Therefore, it is difficult to diagnose lesions using
degradation of the edge region. In spite of minimizing
ultrasound images due to this noise. In this paper, we propose an
algorithm using speckle reducing anisotropic diffusion (SRAD) deterioration in the edge region, these methods are not suitable
and a guided filter for speckle noise reduction and edge for removing speckle noise because they do not take into
preservation. First, speckle, which is multiplicative noise, is account the noise characteristics of the ultrasound image. In
removed by using SRAD. The remaining noise in the filtered image order to overcome the limitation of the single filtering method,
is converted to additive noise using logarithmic transformation. a hybrid filtering method combining various techniques has
The additional noise remaining in the filtered image is further been proposed. Yang et al. [6] removed noise from the
removed using a guided filter. Finally, the noise-free image ultrasound image using a hybrid de-noising method considering
obtained using exponential transform. Experimental results show the local statistics of speckle and NLM filter. Zhang et al. [7]
that the proposed algorithm exhibits an excellent speckle noise
applied the de-speckling method based on the improved
reduction and edge preservation performance than the
conventional filtering techniques. wavelet filter and the guided filter to remove the speckle noise
in the ultrasound image. Coupe et al. [8] removed noise using a
Keywords—Ultrasound image; Speckle noise; Speckle reducing Bayesian framework to derive NLM filter modified for speckle
anisotropic diffusion; Guided filter noise. Despite these efforts, existing research techniques show
low edge preservation and noise reduction performance in the
image, so ultrasound imaging showed low clinical utility.
I. INTRODUCTION Therefore, in the ultrasonic image, a technique for removing
Ultrasound imaging devices are more secure and portable noise and preserving edge information is required.
than other medical diagnostic devices such as X-ray, Computed In this paper, we propose a new algorithm using speckle
Tomography (CT), and Magnetic Resonance Imaging (MRI). reducing anisotropic diffusion (SRAD) [9] and a guided filter
With the development of digital signal processing technology, [10] to effectively remove the speckle noise and preserve the
the resolution of ultrasonic diagnostic devices has been steadily edge region.
improving; however, the image quality of this is lower than The composition of this paper is as follows. In section 2, we
other medical diagnostic devices. The main cause of this describe a noise reduction method using SRAD and guided filter.
problem is the noise of the granular pattern in the acquiring Section 3 compares the proposed algorithm with existing
process of the ultrasound image, which is called speckle noise filtering techniques qualitatively and quantitatively. Section 4
[1]. The speckle noise represents a characteristic of concludes the paper.
multiplicative noise that differs from the general image
obtained through optical sensors, thus it is difficult to remove
the noise [2]. The noise has many difficulties in diagnosing II. PROPOSED ALGORITHM
lesions using ultrasound images, so speckle noise reduction in Figure 1 shows the flowchart of the proposed algorithm in
ultrasound images play an important role in medical image this paper. The proposed algorithm first removes the speckle
processing. noise by applying SRAD to the noise image. Next, the SRAD
Over the last decades, many studies have been conducted to filtering result image is applied to logarithmic transformation
eliminate speckle noise. There are three main methods: linear for transforming remaining multiplicative noise in the image
filtering, Nonlinear filtering, and hybrid filtering methods. into additive noise. The additive noise in the image is further
Linear filtering techniques, such as Gaussian filter and Mean removed using a guided filter. Then, the noise-free image is
filter, can effectively remove noise from the image, but have the obtained through the exponential transformation.
problem of blurring edges. To overcome the problem, nonlinear

978-1-5386-2615-3/18/$31.00 ©2018 IEEE


G
Fig 1. Flowchart of the proposed algorithmG

A. Speckle Reducing Anisotropic Diffusion


AD filter uses the gradient magnitude to distinguish edge regions (a) (b)
in the image [4]. In the edge region of the image, this method preserves Fig 2. Two test images in this study: (a) Cameraman; (b) Airplane.
the edge region using an all-pass filter and removes the noise using a
Gaussian filter in the homogeneous region [9]; however, AD filter where ܽ௞ and ܾ௞ are linear coefficients and this value is a
exhibits an excellent noise reduction performance only in images
constant in the domain ‫ݓ‬௞ . The cost function is used to obtain
degraded by additive noise [9]. SRAD had been modified by Yu and
Action from AD filter to fit the speckle noise generated by the
the coefficients ܽ௞ and ܾ௞ in Eq. (4) [11]. To obtain the linear
ultrasound image [9]. coefficients ܽ௞ and ܾ௞ defined in equation (4), the output image
SRAD uses the instantaneous coefficient of variation (ICOV) to ݃௜ is expressed as follows.
distinguish the edge regions in the image. ICOV is expressed by the
following equation. ݃௜ ൌ ‫݌‬௜ െ  ݊௜ (5)
G
ଶ where ‫݌‬௜ and ݊௜ indicate the input image and the noise of the
ͳ ߘ‫ ܫ‬ଶ ͳ ߘଶ‫ܫ‬
ቀ ቁ ቀ ቁ െ ቀ ଶ ቁ ൬ ൰ image, respectively. The linear coefficients are obtained by Eq.
ʹ ‫ܫ‬ Ͷ ‫ܫ‬
‫ݍ‬ሺ‫ݔ‬ǡ ‫ݕ‬Ǣ ‫ݐ‬ሻ ൌ  ඪ ଶ G (1) (6) which minimizes the cost function of the input image and
ͳ ߘଶ‫ܫ‬ the output image.
൤ͳ ൅  ቀ ቁ ൬ ൰൨
Ͷ ‫ܫ‬
G G
where ‫ ׏‬and ‫׏‬ଶ denote the gradient and the Laplacian, respectively. ሺܽ௞ ǡ ܾ௞ ሻ ൌ ෍ ሺሺ ܽ௞ ‫ܫ‬௞ ൅  ܾ௞ െ ‫݌‬௜ ሻଶ ൅ ߝܽ௞ଶ (6)
ICOV represents the high value in the edge region and low value in the ௜‫א‬௪ೖ
homogeneous region.G
G G where ߝ is a regularization parameter that prevents penalizing
ͳ large ܽ௞ . Thus ߝ preserves the edge region of the image and
ܿሺ‫ݍ‬ሻ ൌ  G
ͳ ൅ ሾ‫ݍ‬ଶ ሺ‫ݔ‬ǡ ‫ݕ‬Ǣ ‫ݐ‬ሻ െ  ‫ݍ‬଴ଶ ሺ‫ݐ‬ሻሿ (2)
ሾ‫ݍ‬଴ଶ ሺ‫ݐ‬ሻ൫ͳ ൅ ‫ݍ‬଴ଶ ሺ‫ݐ‬ሻ൯ሿ
G G
where ‫ݍ‬଴ ሺ‫ݐ‬ሻ is the speckle scale function. The speckle scale function
‫ݍ‬଴ ሺ‫ݐ‬ሻ used in the diffusion coefficient ܿሺ‫ݍ‬ሻ of equation (2) controls the
amount of smoothing in the homogeneous region (Eq. (3)). Therefore,
the speckle scale function removes the noise from the homogeneous
region.
G G
‫ݍ‬଴ ሺ‫ݐ‬ሻ  ൎ  ‫ݍ‬଴ ‡š’ሾെ‫ݐ‬ሿG (3)
G G
തതതതത denote the intensity variance and mean over
where ˜ƒ”ሾœሺ–ሻሿ and œሺ–ሻ (a) (b)
a homogeneous region at G‫ݐ‬S respectively. SRAD processes directly to
maintain useful information without log-compressed data [9], so this
technique is used as a preprocessing filter before applying a
logarithmic transformation.G
G
B. Guided Filter
The guided filter has a filter structure that removes the
speckle noise generated in the ultrasound image using the
guidance image and preserves the edge region. It is assumed that
the guidance image of Eq. (4) can model the output image based
on a linear transform [10]. (c) (d)
Fig 3. Results (Cameraman) of different filtering techniques: (a) Noisy image;
(b) Gaussian filter; (c) Bitonic filter; (d) Proposed algorithm.
݃௜ ൌ  ܽ௞ ‫ܫ‬௞ ൅  ܾ௞ ǡ ‫ݓ  א  ݅׊‬௞ (4)
Table 1 PSNR and SSIM results for Cameraman image
PSNR [dB] SSIM
Noisy 18.62 0.42
BM3D 18.77 0.28
Guided 21.54 0.48
Frost 22.36 0.47
Gaussian 22.50 0.62
Lee 23.14 0.56
Bitonic 25.18 0.74
Proposed 25.98 0.76
(a) (b)

USA).
To evaluate speckle noise reduction and edge preservation
performance of the proposed algorithm, we compared this
method with the conventional filtering techniques such as
Gaussian, Lee [13], Frost [14], block-matching 3D (BM3D) [15],
Guided [16], and Bitonic [17] filters. Peak Signal-to-Noise Ratio
(PSNR) and Structural Similarity Index Measure (SSIM) were
used to quantitatively evaluate speckle noise reduction and edge
preservation performance of the proposed algorithm and
existing filtering techniques.
(c) (d) Tables 1 and 2 show quantitatively experimental results for
Fig 4. Results (Airplane) of different filtering techniques: (a) Noisy image; (b) each image. Table1 exhibited the quantitatively experimental
Gaussian filter; (c) Bitonic filter; (d) Proposed algorithm. results for the Cameraman image. In the Cameraman image, the
proposed algorithm had an excellent noise reduction and edge
removes the noise. To minimize the equation (6), the linear preservation performance (PSNR = 25.98 dB, SSIM = 0.76),
function is obtained by using the cost function. while BM3D filtering technique had the lowest noise reduction
and edge preservation performance with PSNR = 18.77 dB,
ͳ SSIM = 0.28 (Table 1). In the Airplane image of Table 2, the
σ ‫ ܲ ܫ‬െ  ߤ௞ ‫݌‬ҧ௞
ȁ‫ݓ‬ȁ ௜‫א‬௪ೖ ௜ ௜ (7) proposed algorithm had the best noise reduction and edge
ܽ௞ ൌ 
ߪ௞ଶ ൅ ߝ preservation performance with PSNR = 26.55 dB, SSIM = 0.78,
and the guided filtering method had the lowest performance
ܾ௞ ൌ  ‫݌‬ҧ௞ െ ܽ௞ ߤ௞ (8) (PSNR = 19.16 dB, SSIM = 0.28).
Figures 3 and 4 show the results of applying filtering
where ߤ௞ and ߪ௞ଶ denote the mean and variance of the window techniques on Cameraman and Airplane images. Among these
‫ݓ‬௞ in the guidance image. ȁ‫ݓ‬ȁ indicates the numbers of pixels images, the Cameraman image represents the feature including
in the region ‫ݓ‬௞ , and ‫݌‬ҧ௞ denotes the average of the input image many homogeneous regions (Fig. 3). This image such
ܲ௜ in the window ‫ݓ‬௞ . In order to remove the noise in the characteristic applied to the Gaussian filter had a low noise
ultrasound image, the average of the linear coefficients in reduction while an excellent edge preservation performance, but
Equations (7) and (8) is calculated for windows ‫ݓ‬௞ in the the resulting image of the Bitonic filter had a blurring
guidance image. And the average of the linear coefficients is phenomenon as a whole. The proposed algorithm compared to
used in the linear transformation of Eq. (1). This filtering the existing filtering techniques had an excellent speckle noise
technique removes noise in the image using averaging reduction and edge preservation performance. Airplane image
techniques [12]. The output image ݃௜ can be expressed as shows characters that include many edge regions compared to
follows. the existing filtering techniques had an excellent speckle noise
reduction and edge preservation performance. Airplane image
shows characters that include many edge regions compared to
݃௜ ൌ  ܽത௜ ‫ܫ‬௜ ൅ ܾത௜ (9)
Cameraman image (Fig. 4). This image applied the Gaussian
here, ܽത௜ and ܾത௜ mean the average of ܽ௞ and ܾ௞ in the window Table 2 PSNR and SSIM results for Airplane image
‫ݓ‬௞ . PSNR [dB] SSIM
Noisy 16.54 0.21
Guided 19.16 0.28
III. EXPERIMENTAL RESULT BM3D 20.62 0.45
In this study, 8-bit gray level Cameraman (256 x 256 pixels) Frost 22.05 0.36
and Airplane (512 x 512 pixels) standard images were selected Bitonic 22.74 0.69
to evaluate the noise reduction performance (Fig. 2). We added Lee 23.95 0.50
speckle noise to each image. All image processing was Gaussian 25.21 0.66
performed with MATLAB® (R2017a MathWorks®, Natick, MA, Proposed 26.55 0.78
[6] J. Yang, J. Fan, D. Ai, X. Wang, Y. Zheng, S. Tang, and Y. Wang, “Local
filtering method showed a low noise reduction and edge statistics and non-local mean filter for speckle noise reduction in medical
ultrasound image,” Neurocomputing, vol. 195, pp. 88-95, June 2016.
preservation performance. The resulting image of the Bitonic
[7] J. Zhang, G. Lin, L. Wu, and Y. Cheng, “Speckle filtering of medical
filter exhibited an excellent noise reduction performance but ultrasonic images using wavelet and guided filter,” Ultrasonics, vol. 65,
showed a low edge preservation performance. The proposed pp. 177-193, Feb. 2016.
algorithm had an excellent noise reduction and edge [8] P. Coupe, P. Hellier, C. Kervrann, and C. Barillot, “Nonlocal means-based
preservation performance. speckle filtering for ultrasound images,” IEEE Trans. Image Process., vol.
18, no. 10, pp. 2221-2229, Oct. 2009.
[9] Y. Yu and S. T. Acton, “Speckle reducing anisotropic diffusion,” IEEE
IV. CONCLUSION Trans. Image Process., vol. 11, no. 11, pp. 1260-1270, Dec. 2002.
The speckle noise generated in the ultrasound image [10] K. He, J. Sun, and X. Tang, “Guided image filtering,” IEEE Trans.
represents a different character from the conventional noise, so Pattern Anal. Mach. Intell., vol. 35, no. 6, pp. 1397-1409, June 2013.
this feature is difficult to remove speckle noise with existing [11] T. Hastie, R. Tibshirani, and J. H. Friedman, The Elements of Statistical
Learning, Springer, 2003.
filtering techniques. Since speckle noise in ultrasound images
[12] V. Katkovnik, A. Foi, K. Egiazarian, and J. Astola, “From Local Kernel
interferes with the diagnosis of lesions through the image, this to Nonlocal Multiplie-Model Image Denoising,” Int’l . Computer Vision,
noise reduction technique in the ultrasound image is an vol. 86, no. 1, pp. 1-32, 2010.
important technology in medical image processing. [13] J. S. Lee, “Digital image enhancement and noise filtering by use of local
In this paper, we proposed a new algorithm that used SRAD statistics,” IEEE Trans. Pattern Anal. Mach. Intell., vol. PAMI-2, no. 2,
and guided filter considering noise characteristics to reconstruct pp. 165-168, Mar. 1980.
images degraded by speckle noise in the ultrasound image. [14] V. S. Frost, J. A. Stiles, K. S. Shanmugan, and J. C. Holtzman, “A model
Speckle noise in the ultrasound image transformed for radar images and its application to adaptive digital filtering of
multiplicative noise into additive noise using logarithmic multiplicative noise,” IEEE Trans. Pattern Anal. Machine Intell., vol.
PAMI-4, pp. 157-165, 1982.
transformation. First, this noise was eliminated by ‫ێې‬ing a filter
[15] K. Dabov, R. Foi, V. Katkovnik, and K. Egiazarian, “Image Denoising by
(SRAD) that could be directly applied to multiplicative noise. Sparse 3D Transform-Domain Collaborative Filtering,” IEEE Trans.
The remaining noise in the SRAD result image converted Image Process., vol. 16, no. 8, pp. 2080-2095, Aug. 2007.
characteristics (additive noise) of the speckle noise using [16] K. He, J. Sun, and X. Tang, “Guided image filtering,” IEEE Trans. Pattern
logarithmic transformation. Next, the guided filter was applied Anal. Mach. Intell., vol. 35, no. 6, pp. 1397-1409, June 2013.
to the SRAD result image to additionally remove the remaining [17] G. Treece, "The Bitonic Filter: Linear Filtering in an Edge-Preserving
additive noise in the image. Experimental results exhibited that Morphological Framework," IEEE Trans. Image Process., vol. 25, no. 11,
pp. 5199-211, Nov 2016.
the proposed algorithm had represented an excellent noise
reduction and edge preservation performance than the existing
filtering techniques. In future research, we will study the guided
filter which can be directly applied multiplicative noise.

ACKNOWLEDGMENT
This research was supported by Basic Science Research
Program through the National Research Foundation of
Korea(NRF) funded by the Ministry of Science, ICT and future
Planning(NRF-2015R1A2A2A01006004)

REFERENCES

[1] J. G. Abbott and F. L. Thurstone, “Acoustic speckle: Theory and


experimental analysis,” Ultrasonic Imaging, vol. 1, no. 4, pp. 303-324,
1979.
[2] H. Kato and J. W. Goodman, "Nonlinear filtering of coherent optical
system through half-tone screen processes,” Appl. Opt., vol. 14, No. 8, pp.
1813-1824, 1975.
[3] A. Buades, B. Coill, and J. -M. Morel, “A non-local algorithm for image
denoising,” in Proc. Comput. Soc. Conf. Comput. Vis. Pattern Recognit.,
vol. 2. pp. 60-65, 2005.
[4] P. Perona and J. Malik, “Scale-space and edge detection using anisotropic
diffusion,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 12, no. 7, pp.
629-639, 1990.
[5] C. Tomasi and R. Manduchi, “Bilateral filtering for gray and color images,”
in Proc. Int. Conf. Comput. Vis., pp. 839-846, 1998.

You might also like