Second order total generalized variation for speckle reduction in ultrasound images
Introduction
Ultrasound imaging is widely employed in diagnosis of the clinical disease due to its low cost, safety, noninvasive measuring and real-time imaging [1], [2]. Unfortunately, the speckle noise in ultrasound images influences the diagnosis accuracy of the pathological and normal tissues. Speckle noise is a granular pattern based on the scattered ultrasound beam from the body’s tissues, which degrades the image contrast and resolution. Thus, it is a daunting challenge to eliminate the speckle noise in ultrasound images and simultaneously retain the important features.
As far as we know, the speckle noise in coherent imaging system follows Gamma distribution [3], [4]. However, the speckle noise in ultrasound images does not satisfy Gamma distribution due to the different imaging theory, see [5], [6], [7] and references therein. According to [5], [8], the degradation model of ultrasound images is formulated as the following form: where u > 0 is the underlying image on a rectangle u0 is the degraded image, and η denotes the Gaussian noise with mean 0 and standard deviation σ. We aim at recovering u from its noisy observation u0.
In recent decades, numerous denoising methods have been applied to recover medical ultrasound images corrupted by the speckle noise. The major methods are divided in three categories, including locally adaptive statistic filters [9], [10], [11], [12], wavelet transform based thresholding methods [13], [14], [15] and variational methods [16], [17], [18], [19], [20], [21], [22], [23]. In [17], motivated by the classical ROF model [24], the authors proposed a convex variational model for removing the speckle noise in ultrasound images: where is referred as the TV regularization term, λ > 0 represents the regularization parameter which compromises between the data fidelity term and the TV regularization term. They utilized the gradient projection method to solve the model (2) and gave the relevant theoretical analysis. In [17], the regularization parameter λ is adaptively updated based on the estimated standard variance of the noise σ. But it was hard to accurately estimate σ. In [18], Huang et al. applied a generalized Kullback-Leibler (KL) distance [25], [26] to measure the discrepancy between u and u0. Their variational model was given as follows: Since the above model is nonconvex, they took and regarded z as an image in the logarithm domain. Thus, the minimization model for removing the speckle noise in ultrasound images was rewritten as Note that as u0 > 0, the objective function of Eq. (3) is strictly convex. They solved this model by the split Bregman iterative method. Finally, the denoising result is attained by .
The TV regularization effectively reduce the noise and preserve the sharp edges. But there still exists some undesired staircase effect in smooth regions of the restored images. However, the TGV regularization is a useful tool to remove the staircase effect as well as preserve the sharp edge. In [27], [28], [29], [30], the researches show that TGV regularization involves the higher-order derivatives of images, which restores fairly accurate pixels in homogeneous regions and wipe off the staircase effect. Fig. 1 shows the denoising performance of the TV and TGV regularization for the noisy image corrupted by the additive Gaussian noise. From Fig. 1, we find that the TGV regularization accurately describes the intensity variations in homogeneous regions while preserves the edges.
Recently, most literatures have studied the TGV regularization to remove the additive Gaussian noise [31], [32] or the multiplicative noise [33], [34]. The TGV regularization substantially reduce the staircase effect, while effectively remove the noise and preserve the sharp edges. Motivated by the property of the TGV regularization, we propose a new convex model which consists of the second-order TGV regularization and the data fidelity term . Furthermore, the ADMM algorithm is developed to solve the proposed model. The numerical experiments demonstrate the efficiency and effectiveness of the proposed model.
The remainder of this paper is organized as follows. In Section 2, the TGV regularization and the ADMM algorithm are reviewed briefly. Then we propose a novel variational model to reduce the speckle noise in ultrasound images and prove the existence of the solutions in Section 3. Meanwhile, the ADMM algorithm is developed for solving the proposed model. In Section 4, numerical results show that the performance of our proposed model is better than some existing approaches. Finally, the conclusions are given in Section 5.
Section snippets
A review of the TGV regularization and ADMM
In this section, we give a short review about the TGV regularization and the ADMM algorithm for the sake of completeness.
The proposed model based on
According to [8], [29], we propose a new variational model which combines the data fidelity term with the second-order TGV regularization for the speckle reduction. Consequently, the proposed denoising model is given by where λ is the positive regularization parameter.
In order to illustrate the existence of solutions for Eq. (7), we firstly consider the general minimization problem: where p ∈ (1, ∞), and
Experiments
In this section, we present several experimental results on both synthetic and real ultrasound images, respectively. All of numerical experiments are performed under Windows 7 and Matlab Version 7.10 (R2012a) running on a desktop with 3.4 GHz Intel Core i3-2130 CPU and 4 G RAM memory. In order to demonstrate the efficiency and effectiveness of the proposed TGV-based model, we give the sufficient experiments by compared with the TVGD model [17], the TVADM model, the TVKL model [18] and the TGVPD
Conclusion
In this paper, we propose a new variational model based on the TGV regularization. The new model effectively eliminates the speckle noise in ultrasound image as well as alleviates the staircase effect when utilizing the TV regularization. We develop the ADMM algorithm to solve the proposed model. The numerical experiments illustrate the efficiency and effectiveness of the proposed model. Compared with some state-of-the-art models, the proposed model achieves the visual and quantitative
Acknowledgments
We would like to thank the authors of [17] for providing the codes and the authors of [53] for offering the real ultrasound images. This research is supported by NSFC (61772003, 61402082, 11401081) and the Fundamental Research Funds for the Central Universities (ZYGX2016J129).
References (53)
- et al.
Lmmse and map estimators for reduction of multiplicative noise in the nonsubsampled contourlet domain
Signal Process.
(2009) - et al.
Despeckling of medical ultrasound images using daubechies complex wavelet transform
Signal Process.
(2010) - et al.
Image restoration using total variation with overlapping group sparsity
Inf. Sci.
(2015) - et al.
Fast reduction of speckle noise in real ultrasound images
Signal Process.
(2013) - et al.
Generalized total variation-based MRI RICIAN denoising model with spatially adaptive regularization parameters
Magn. Reson. Imag.
(2014) - et al.
Stripe noise removal of remote sensing images by total variation regularization and group sparsity constraint
Remote Sens.
(2017) - et al.
Tensor completion using total variation and low-rank matrix factorization
Inf. Sci.
(2016) - et al.
Nonlinear total variation based noise removal algorithms
Phys. D Nonlinear Phenom.
(1992) - et al.
Denoising optical coherence tomography using second order total generalized variation decomposition
Biomed. Signal Process. Control
(2016) - et al.
High-order tvl1-based images restoration and spatially adapted regularization parameter selection
Comput. Math. Appl.
(2014)
Group sparsity based regularization model for remote sensing image stripe noise removal
Neurocomputing
Group-based image decomposition using 3-d cartoon and texture priors
Inf. Sci.
Image deblurring with an inaccurate blur kernel using a group-based low-rank image prior
Inf. Sci.
A new speckle filtering method for ultrasound images based on a weighted multiplicative total variation
Signal Process.
Diagnostic Ultrasound Imaging: Inside Out
Introduction to Biomedical Imaging
A variational approach to removing multiplicative noise
SIAM J. Appl. Math.
Removing multiplicative noise by Douglas–Rachford splitting methods
J. Math. Imag. Vis.
Adaptive speckle reduction filter for log-compressed b-scan images
IEEE Trans. Med. Imag.
On the statistical characteristics of log-compressed rayleigh signals: theoretical formulation and experimental results
J. Acoust. Soc. Am.
Speckle-constrained filtering of ultrasound images
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
An adaptive weighted median filter for speckle suppression in medical ultrasonic images
IEEE Trans. Circ. Syst.
Speckle reducing anisotropic diffusion
IEEE Trans. Image Process.
Oriented speckle reducing anisotropic diffusion
IEEE Trans. Image Process.
Nonlocal means-based speckle filtering for ultrasound images
IEEE Trans. Image Process.
Cited by (20)
Impulse noise removal by using a nonconvex TGV regularizer and nonconvex fidelity
2023, Journal of the Franklin InstituteHybrid non-convex regularizers model for removing multiplicative noise
2022, Computers and Mathematics with ApplicationsAn adaptive total generalized variational model for speckle reduction in ultrasound images
2022, Journal of the Franklin InstituteCitation Excerpt :Recently, the variational methods for speckle reduction in ultrasound images have been extensively studied. In the literature, there are mainly two categories of denoising methods for ultrasound image restoration, namely, total variation (TV) based methods [1,14,15,27] and higher order TV based methods [10,17,23,34]. In the TV based approach, the denoising methods generally involve two terms.
Generalized singular value thresholding operator based nonconvex low-rank and sparse decomposition for moving object detection
2019, Journal of the Franklin InstituteCitation Excerpt :In this paper, we directly consider a general nonconvex penalty function onto the singular values of the low-rank matrix and propose a generalized formulation for nonconvex low-rank and sparse decomposition (NonLRSD) problem. And then, we solve the proposed nonconvex problem by the alternating direction method of multipliers (ADMM) [33–37]. During the solving process, we use the generalized singular value thresholding (GSVT) operator [38] instead of the singular value thresholding (SVT) operator [39] to solve the proposed nonconvex low-rank minimization problem.
A new wavelet family for speckle noise reduction in medical ultrasound images
2019, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :Filters for images can also be characterized as linear or non-linear. Some linear filters have been proposed in the literature: as the minimization model based on second order total variation [16], an hybrid local guided and nonlocal mean method [17], filter using binary local standards [18] and the Gaussian filter of the wavelet coefficients [10] for speckle reduction. However most researchers have been interested in non-linear filters because of their edge-preserving and smoothing techniques.