WorksEngineering
Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis
Abstract
Robust Principal Component Analysis (RPCA) is essential in signal processing, with applications in hyperspectral imaging, face recognition, and video background modeling. Traditional RPCA methods often suffer from insufficient utilization of data priors and struggle to capture complex image structures. Conversely, advanced deep denoisers like DnCNN, FFDNet, and DRUNet excel at leveraging image data priors but lack physical model constraints and demonstrate poor robustness against structured noise in RPCA tasks. This paper proposes a novel Plug-and-Play (PnP) method for multi-channel RPCA, integrating both the physical generative model and cutting-edge deep priors. Extensive experiments on RGB and hyperspectral datasets show that PnP-PDA achieves better quantitative performance than conventional RPCA methods and state-of-the-art methods, including deep denoisers and tensor RPCA approaches, while preserving visual details.
Introduction
ROBUST principal component analysis (RPCA) is a fundamental topic in signal processing, with numerous applications in hyperspectral imaging [1], robust face recognition [2], and video background modeling [3]. In RPCA, an image is typically decoupled into two components: a low-rank matrix representing the background, and a sparse object, which may be a target or structured noise. Based on the physical model, RPCA has been utilized for principal component pursuit (PCP) for matrix low-rank recovery [4], stable PCP (SPCP) for noise-corrupted observations [5], outlier pursuit for column-sparse corruptions [6], and multi-channel RPCA [7], [8]. However, conventional RPCA often makes limited use of data priors, and the single low-rank structure cannot capture the intricate prior details of the images.
On the other hand, recent advances in deep learning have led to powerful deep denoisers for image denoising, including DnCNN [9], FFDNet [10], and DRUNet [11], to name a few. These denoisers are well-trained to capture the data priors of images and have demonstrated remarkable noise removal effects. However, they often lack physical model information and are less robust to structured noise, which can lead to degraded performance in RPCA tasks when faced with sparse impulse noise and heavily corrupted observations.
More recently, the trend of combining physical model methods with emerging deep priors has created a new paradigm for handling complex tasks in computational imaging. In this context, plug-and-play (PnP) and regularization by denoising (RED) have garnered increasing attention. For instance, PnP priors [11], [12] replace model-based proximal operators with pre-trained denoisers to recover fine textures. RED [13] constructs an explicit regularizer from the denoiser output, providing a variational interpretation that complements the implicit PnP formulation. While these approaches have been applied to various computational imaging tasks, including denoising, super-resolution, and deblurring, they have not yet been explored for more challenging RPCA tasks.
This letter focuses on leveraging the PnP framework for multi-channel RPCA, aiming to combine both physical models and deep priors. We propose an RPCA formulation that is implicitly regularized by data priors driven by deep denoisers. We apply the nonconvex norm for the structured noise removal and Schatten- norm for promoting the low-rank background for better separation. For the resulting problem, we introduce a PnP method embedded in a dedicated proximal distance algorithm (PDA). PnP-PDA decouples physical-model optimization and prior regularization, and uses Nesterov extrapolation for acceleration. We conducted extensive experiments on thirteen datasets, including RGB images and hyperspectral image (HSI) datasets. The results show that PnP-PDA achieves better recovery performance than conventional RPCA methods and the compared state-of-the-art deep denoisers and tensor RPCA approaches.
Background and Problem Formulation
Background of RPCA
Consider the following robust recovery signal model
where represents the noise corrupted multi-channel observation, denotes the low-rank background component, denotes the fiber-sparse outlier noise, and denotes the additive Gaussian noise. Here, denotes the data spatial size and denotes the number of channels.
The objective of robust multi-channel recovery is to separate and estimate and from . This model arises in tasks such as multispectral imaging [1] and video surveillance [7], where captures the stationary scene or spectral background and captures small objects or anomalous pixels.
We first review the conventional RPCA formulation for the single-channel case, i.e., in (1), which is given by
where denotes the low-rank component in the matrix case, and denotes the sparse outliers, balances the two terms, and is the nuclear norm. This problem can be solved by accelerated proximal gradient (APG) [14], alternating direction method of multipliers (ADMM) [15], or singular value thresholding (SVT) [16].
Beyond (2), the principles of RPCA have been exploited for low-rank and sparse decomposition, including SPCP for observations with mixed Gaussian noise [5], [17], tensor decomposition methods [7], [18]. However, these conventional methods remain an issue: they often rely on physical models and typically explore less on the data priors, which may cause prior mismatch and limited recovery under severe corruptions.
Problem Formulation
To better harness both the physical model and data priors, we formulate a PnP RPCA formulation with an implicit deep prior, which reads as
In this formulation, is an implicit deep prior, and balance the four regularizers. The norm for enforces fiber-sparsity along the mode-3 fibers . The tensor Schatten- norm is , where is the matrix Schatten- norm with being the -th singular value of . Here, denotes the tensor mode- matricization with dimensions , and , while the inverse operator restores the multi-channel shape. The anisotropic TV norm is expressed in a channel-wise vectorized form as . Here, denotes the vectorization of the -th spatial channel, and are the horizontal and vertical finite-difference matrices evaluated under periodic boundary conditions.
Compared with (2), the formulation in (3) integrates the physical model with a deep prior through the PnP framework. Furthermore, to better promote low-rank and sparsity structures, we replace the nuclear norm and norm with the nonconvex Schatten- and norms. However, the enhanced PnP formulation (3) presents a critical nonconvex optimization challenge, particularly due to the coupling between the deep prior and the other components.
Plug-and-Play Proximal Distance Algorithm
PnP-PDA
We address the nonconvex coupling challenge with a novel PDA algorithm that reformulates the task as a consensus optimization problem. Then, the global objective can be split into tractable subproblems that allow for efficient closed-form updates. This decoupling isolates the deep prior , enabling the direct embedding of a deep denoiser as a modular solver.
Our first step is to introduce the consensus constraint with by
.
Then, problem (3) is reformulated as
where . Next, we transform the consensus constraint with a soft constraint, which is posed as
where is the distance function and is a penalty parameter. We majorize at using a quadratic surrogate
with equality at , where denotes the projection onto that takes the form with
Under this majorization, we propose the following PDA algorithm where at iteration it takes the form
Since is separable, problem (8) splits into four independent subproblems for , and , each admitting a form of proximal operator and can be efficiently solved.
Proximal Mapping for : The subproblem reads
which amounts to derive the proximal mapping . Since is an implicit data prior, its proximal subproblem can be addressed by a denoiser under the PnP framework [19]:
where is the noise level parameter.
Proximal Mapping for : The subproblem is
Since the subproblem sums over three mode unfoldings coupled through the same tensor variable, we approximate it using a mode-wise decoupling-and-averaging strategy with three matrix subproblems for , written as
Let be the SVD of the mode- matricization of , where denotes its -th singular value. By the unitary invariance of the Schatten- and Frobenius norms, the optimal shares the singular vectors and of [16], and the optimization reduces to independent scalar problems for the sought singular values
Since , the term is non-convex. Following the reweighted least squares (RLS) strategy [20], we majorize this term at the current iterate by the first-order upper bound , where is a constant and the scalar weight is
with denoting the corresponding singular value of the previous iterate and ensuring numerical stability. Substituting this majorizer back into (13) and setting the derivative to zero yields the closed-form reweighted shrinkage
Averaging the three mode estimates and refolding yields the multi-channel update
where . Within each PDA iteration, the RLS performs a single reweighting step.
Proximal Mapping for
The -subproblem is
where denote the -th channel of and , respectively. Since the problem is independent across channels, we describe the procedure for each channel. Since and couple the entries of , we adopt Fast Total Variation deconvolution (FTVd) [21] by introducing auxiliary variables and and relaxing the equality constraints with a quadratic penalty of weight ,
We choose . For any given , the solutions for and is given by soft-thresholding
where . Substituting back into (18) and setting the gradient with respect to to zero gives the normal equation , where
Under periodic boundary conditions, and are circulant and diagonalized by the DFT matrix as and . Then, the closed-form solution is
Algorithm 1 PnP-PDA for RPCA
Require: Observation , parameters , smoothing factor , scaling factor , and .
Initialize: ; ; .
repeat
Compute extrapolated points via (23);
Compute consensus projection via (7);
Update via deep denoiser (10);
Update via RLS singular value shrinkage (16);
Update via TV proximal operator (20);
Update via RLS tubal shrinkage (22);
Update penalty: ;
until convergence
Return .
where is a constant diagonal matrix independent of and , precomputed once and reused across all channels.
Proximal Mapping for
The subproblem reads as
Applying the same RLS scheme as in the subproblem yields the closed-form reweighted tubal shrinkage
where is the fiber-wise weight at spatial location .
PDA Acceleration and Complexity Analysis
We introduce Nesterov-type extrapolation to accelerate numerical convergence, which has been shown to effectively accelerate convergence in many optimization algorithms [22, 23, 24]. At iteration , the extrapolated point is calculated by
and similarly for . The PDA update is then performed at the extrapolated point by replacing with and with in the consensus projection (7), while the four proximal subproblems (10)–(22) remain unchanged. We also adopt a homotopy strategy for by initializing it with a small value and increasing it via with , which gradually enforces the consensus constraints. The complete procedure is summarized in Algorithm 1. The algorithm complexity analysis is provided in the supplementary material.
Simulation Results
Simulation Settings and Benchmarks
We evaluate PnP-PDA on thirteen images, including ten RGB images and three HSI datasets: PaviaU (103 bands), Samson (156 bands), and Jasper (198 bands). All data are tested under two corruption ratios (10% and 20%) and Gaussian noise (). Corrupted fibers are perturbed by i.i.d. uniform noise in . The PnP framework can plug different pre-trained denoisers for different datasets, including DRUNet [11] for RGB images and DPHSIR [27] for hyperspectral images. We use PSNR, SSIM [28], and LPIPS [29] metrics for RGB images, and band-averaged MPSNR, MSSIM [30] metrics for HSI. The definitions of these metrics are provided in the Information for Reproducibility.
We compare PnP-PDA against seven benchmarks and one ablation baseline. Two are conventional RPCA methods: Inexact ALM (IALM) [15] and SPCP [5]. Three are state-of-the-art deep denoisers: DnCNN [9], FFDNet [10], and DRUNet [11]. Two are state-of-the-art tensor RPCA methods: Unfolded ScaledGD (U-SGD) [25] and t-CTV [26]. We added an ablation study as a baseline by examining the deep learning prior in our PnP-PDA framework.
Performance Comparison
Figure 1 compares results on Girl and Bell at . Conventional RPCA methods, the ablation variant, and state-of-the-art tensor RPCA methods produce over-smoothed results, and deep denoisers leave structured artifacts, whereas PnP-PDA removes corruptions while preserving textures and edges. Extensive RGB and HSI visual results under a wider range of noise levels are provided in the supplementary material.

Figure 1. Visual results on “Girl” and “Bell”: Original, Corruption, IALM, SPCP, DnCNN, FFDNet, DRUNet, U-SGD, t-CTV, Ablation, and PnP-PDA.
Table I reports the results on ten diverse RGB images at . PnP-PDA achieves better performance than the state-of-the-art tensor RPCA baselines across all tests. The improvement over the ablation (without PnP) shows the contribution of the deep prior in our framework. Overall, PnP-PDA demonstrates consistently competitive quantitative performance on all ten RGB images.
| Method (PSNR/SSIM/LPIPS) | Peppers | Girl | Bell | House | Plane | Child | kodim23 | kodim24 | kodim04 | kodim15 | |
| IALM [15] | 0.1 | 13.27/0.3178/0.9547 | 14.94/0.3883/0.8013 | 16.95/0.2421/1.0123 | 14.66/0.3353/0.9113 | 16.17/0.3885/0.7972 | 17.41/0.3971/0.8834 | 14.11/0.3169/1.1248 | 17.20/0.5077/0.6659 | 16.14/0.3486/0.9963 | 15.35/0.3772/0.8045 |
| 0.2 | 12.94/0.2770/1.0247 | 14.04/0.3221/0.9057 | 15.43/0.1898/1.1019 | 13.71/0.2784/1.0148 | 14.60/0.3243/0.9332 | 16.74/0.3401/0.9614 | 13.48/0.2616/1.1699 | 15.91/0.4171/0.8198 | 15.24/0.2869/1.1016 | 14.27/0.3227/0.8925 | |
| SPCP [5] | 0.1 | 12.49/0.2929/1.0147 | 14.67/0.3766/0.8182 | 16.66/0.2342/1.0262 | 14.38/0.3296/0.9140 | 16.04/0.3846/0.8061 | 16.67/0.3767/0.8993 | 13.61/0.2952/1.1334 | 16.49/0.4908/0.6871 | 15.95/0.3447/1.0100 | 15.02/0.3712/0.8155 |
| 0.2 | 12.22/0.2547/1.0872 | 13.82/0.3129/0.9206 | 15.22/0.1857/1.1095 | 13.50/0.2751/1.0209 | 14.47/0.3210/0.9408 | 16.05/0.3245/0.9810 | 13.06/0.2457/1.1771 | 15.37/0.4059/0.8329 | 15.05/0.2839/1.1087 | 13.97/0.3175/0.9002 | |
| DnCNN [9] | 0.1 | 25.83/0.7587/0.1369 | 27.26/0.8013/0.1206 | 29.94/0.8274/0.1709 | 28.19/0.8090/0.0651 | 26.21/0.7953/0.1242 | 25.65/0.6860/0.2440 | 27.02/0.7730/0.1141 | 25.70/0.7694/0.1341 | 27.29/0.7603/0.1402 | 25.99/0.7419/0.1792 |
| 0.2 | 24.26/0.7060/0.1795 | 25.83/0.7539/0.1490 | 28.48/0.7817/0.2113 | 26.56/0.7633/0.1052 | 24.60/0.7320/0.1735 | 23.75/0.6002/0.3364 | 25.51/0.7182/0.1860 | 24.49/0.7100/0.1896 | 25.71/0.7059/0.2093 | 23.98/0.6706/0.2347 | |
| FFDNet [10] | 0.1 | 23.42/0.6101/0.3313 | 23.66/0.6110/0.3404 | 23.93/0.4312/0.5661 | 23.81/0.5495/0.3103 | 23.63/0.6252/0.2765 | 23.38/0.5650/0.4761 | 23.66/0.6020/0.3518 | 22.90/0.6522/0.3123 | 23.37/0.5614/0.4574 | 23.61/0.6043/0.3600 |
| 0.2 | 19.93/0.4429/0.5892 | 19.78/0.4293/0.5413 | 19.70/0.2493/0.8837 | 19.76/0.3692/0.5891 | 19.73/0.4347/0.5267 | 19.75/0.3754/0.8082 | 19.88/0.4289/0.5759 | 19.59/0.4813/0.6036 | 19.65/0.3747/0.7592 | 19.87/0.4157/0.5901 | |
| DRUNet [11] | 0.1 | 21.48/0.5332/0.3694 | 21.45/0.5049/0.3716 | 21.52/0.3157/0.6091 | 21.61/0.4552/0.3375 | 21.58/0.5366/0.3167 | 21.30/0.4719/0.5382 | 21.36/0.5012/0.4031 | 20.87/0.5453/0.3397 | 21.27/0.4580/0.4916 | 21.63/0.5217/0.4266 |
| 0.2 | 18.10/0.3684/0.5966 | 17.77/0.3425/0.5523 | 17.66/0.1809/0.8780 | 17.84/0.2994/0.5787 | 18.06/0.3655/0.5360 | 17.96/0.2969/0.8167 | 17.86/0.3418/0.5814 | 17.50/0.3735/0.6331 | 17.68/0.2852/0.7559 | 18.18/0.3461/0.6105 | |
| U-SGD [25] | 0.1 | 22.50/0.5568/0.3682 | 24.51/0.6560/0.4229 | 26.56/0.5751/0.5182 | 23.37/0.4939/0.3723 | 22.04/0.5413/0.3491 | 24.05/0.6072/0.5059 | 22.65/0.5482/0.3910 | 20.88/0.5784/0.3801 | 22.82/0.5406/0.5515 | 22.43/0.551? |
| 0.2 | 21.?? | 22.?? | 25.?? | 22.?? | 20.?? | 23.?? | 21.?? | 20.?? | 21.?? | 21.?? | |
| t-CTV [26] | 0.1 | 24.36/0.6355/0.2488 | 26.00/0.6766/0.2269 | 26.67/0.5196/0.4352 | 25.32/0.5662/0.2251 | 23.46/0.5930/0.2884 | 25.68/0.6352/0.3220 | 24.33/0.6355/0.2692 | 22.58/0.6779/0.2964 | 24.45/0.6068/0.3148 | 24.09/0.6124/0.2699 |
| 0.2 | 23.69/0.6019/0.2765 | 25.01/0.6339/0.2847 | 25.78/0.4750/0.4940 | 24.50/0.5540/0.2675 | 22.81/0.5559/0.3178 | 24.89/0.5974/0.3819 | 23.66/0.6022/0.3124 | 21.97/0.6403/0.3397 | 23.76/0.5691/0.3645 | 23.54/0.5831/0.2941 | |
| Without PnP (ablation) | 0.1 | 24.21/0.6155/0.2527 | 26.17/0.6959/0.2855 | 27.48/0.6481/0.4630 | 25.02/0.5659/0.2536 | 23.46/0.5871/0.2919 | 25.34/0.6232/0.3345 | 24.20/0.6324/0.3044 | 22.54/0.6601/0.2698 | 24.46/0.6104/0.3363 | 23.89/0.5906/0.2759 |
| 0.2 | 23.48/0.5976/0.3335 | 25.61/0.6827/0.3261 | 27.12/0.6532/0.4942 | 24.44/0.5576/0.2798 | 22.84/0.5710/0.3470 | 24.68/0.6032/0.4180 | 23.64/0.6183/0.3504 | 21.96/0.6304/0.3121 | 23.96/0.5941/0.4052 | 23.11/0.5755/0.3410 | |
| PnP-PDA | 0.1 | 28.09/0.8358/0.0492 | 30.36/0.8654/0.0573 | 32.75/0.9171/0.0364 | 31.02/0.8485/0.0275 | 27.01/0.8442/0.0502 | 29.00/0.8328/0.0590 | 28.29/0.8382/0.0513 | 24.96/0.7900/0.1096 | 27.99/0.7869/0.1330 | 28.30/0.8271/0.0996 |
| 0.2 | 27.01/0.8190/0.0543 | 29.55/0.8558/0.0800 | 31.47/0.9008/0.0461 | 29.76/0.8336/0.0307 | 25.46/0.8248/0.0732 | 27.84/0.8126/0.0897 | 26.80/0.8155/0.0712 | 23.80/0.7264/0.1068 | 26.91/0.7605/0.1299 | 26.95/0.8099/0.0975 |
Table I. Comparison of recovery methods on ten RGB benchmark images at under 10% and 20% corruption.
Table II reports the results on three HSI datasets with 103, 156, and 198 spectral bands. PnP-PDA achieves the highest MPSNR and MSSIM in all 12 test conditions. It consistently outperforms the conventional RPCA methods, state-of-the-art deep denoisers, and other RPCA methods, and the ablation variant. These results support the effectiveness of PnP-PDA for HSI recovery.
| Method (MPSNR/MSSIM) | , PaviaU 0.10 | PaviaU 0.20 | Samson 0.10 | Samson 0.20 | Jasper 0.10 | Jasper 0.20 | , PaviaU 0.10 | PaviaU 0.20 | Samson 0.10 | Samson 0.20 | Jasper 0.10 | Jasper 0.20 |
| IALM [15] | 27.86/0.769 | 25.22/0.670 | 30.98/0.795 | 28.34/0.707 | 29.62/0.823 | 27.05/0.737 | 26.18/0.692 | 24.33/0.621 | 29.30/0.735 | 27.44/0.670 | 27.87/0.765 | 26.06/0.696 |
| SPCP [5] | 27.85/0.768 | 25.17/0.667 | 30.92/0.794 | 28.25/0.704 | 29.60/0.822 | 27.03/0.737 | 26.15/0.691 | 24.27/0.618 | 29.24/0.733 | 27.35/0.667 | 27.84/0.765 | 26.04/0.696 |
| DnCNN [9] | 25.80/0.647 | 24.25/0.619 | 25.48/0.568 | 23.48/0.580 | 25.37/0.614 | 23.52/0.580 | 23.73/0.543 | 23.81/0.551 | 23.70/0.455 | 22.02/0.499 | 23.38/0.522 | 22.21/0.518 |
| FFDNet [10] | 27.21/0.701 | 24.81/0.658 | 28.09/0.752 | 23.94/0.659 | 26.93/0.674 | 23.97/0.627 | 26.02/0.695 | 23.16/0.576 | 26.05/0.714 | 22.43/0.580 | 25.63/0.661 | 22.48/0.559 |
| DRUNet [11] | 27.58/0.730 | 24.44/0.629 | 28.28/0.760 | 23.82/0.628 | 27.18/0.692 | 23.64/0.595 | 25.21/0.650 | 25.14/0.579 | 25.06/0.681 | 22.61/0.611 | 25.04/0.625 | 22.58/0.536 |
| U-SGD [25] | 30.35/0.795 | 26.85/0.627 | 29.18/0.621 | 25.86/0.460 | 30.20/0.774 | 26.95/0.638 | 29.12/0.747 | 26.12/0.589 | 27.14/0.537 | 24.72/0.415 | 28.11/0.704 | 25.73/0.589 |
| t-CTV [26] | 27.42/0.665 | 22.80/0.424 | 28.80/0.621 | 24.18/0.396 | 28.19/0.702 | 23.37/0.474 | 26.58/0.625 | 23.23/0.395 | 28.05/0.585 | 23.62/0.370 | 27.33/0.665 | 22.76/0.444 |
| Ablation | 29.69/0.822 | 22.15/0.395 | 29.80/0.789 | 26.26/0.552 | 30.80/0.840 | 23.85/0.500 | 28.89/0.789 | 21.06/0.344 | 29.28/0.778 | 24.31/0.440 | 29.85/0.812 | 22.40/0.427 |
| PnP-PDA | 31.34/0.864 | 29.38/0.812 | 33.86/0.876 | 29.77/0.820 | 31.33/0.852 | 28.08/0.757 | 30.46/0.830 | 28.80/0.793 | 32.24/0.836 | 28.83/0.795 | 30.09/0.804 | 27.54/0.752 |
Table II. Comparison of HSI recovery methods under 10% and 20% corruption at Gaussian noise levels and .
Conclusion
This letter presents a PnP method for RPCA that integrates both the physical model and deep priors. Our proposed PnP-PDA algorithm transforms the problem into subproblems for evaluating proximal operators, utilizing closed-form updates and deep denoiser processing. Extensive experiments on diverse RGB and HSI datasets demonstrate the efficacy of our PnP-PDA method.
References
- [1]M. M. Salut and D. V. Anderson, “Tensor robust CUR for compression and denoising of hyperspectral data,” IEEE Access, vol. 11, pp. 77 492–77 505, 2023.DOI
- [2]J. Wright, A. Y. Yang, A. Ganesh, S. S. Sastry, and Y. Ma, “Robust face recognition via sparse representation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 31, no. 2, pp. 210–227, 2009.DOI
- [3]Z. Han, S. Zhang, Z. Liu, Y. Wang, J. Yao, and Y. Wang, “Tensor robust principal component analysis with side information: Models and applications,” IEEE Trans. Circuits Syst. Video Technol., vol. 33, no. 8, pp. 3713–3725, 2023.DOI
- [4]E. J. Candès, X. Li, Y. Ma, and J. Wright, “Robust principal component analysis?” J. ACM, vol. 58, no. 3, pp. 1–37, 2011.arxiv.org/abs/0912.3599
- [5]Z. Zhou, X. Li, J. Wright, E. J. Candès, and Y. Ma, “Stable principal component pursuit,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT), 2010, pp. 1518–1522.arxiv.org/abs/1001.2363
- [6]H. Xu, C. Caramanis, and S. Sanghavi, “Robust PCA via outlier pursuit,” IEEE Trans. Inf. Theory, vol. 58, no. 5, pp. 3047–3064, 2012.arxiv.org/abs/1010.4237
- [7]C. Lu, J. Feng, Y. Chen, W. Liu, Z. Lin, and S. Yan, “Tensor robust principal component analysis with a new tensor nuclear norm,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 42, no. 4, pp. 925–938, 2019.DOI
- [8]B. Huang, C. Mu, D. Goldfarb, and J. Wright, “Provable low-rank tensor recovery,” Optimization Online, Tech. Rep. 4252, 2014.
- [9]K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, “Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising,” IEEE Trans. Image Process., vol. 26, no. 7, pp. 3142–3155, 2017.DOI
- [10]K. Zhang, W. Zuo, and L. Zhang, “FFDNet: Toward a fast and flexible solution for CNN-based image denoising,” IEEE Trans. Image Process., vol. 27, no. 9, pp. 4608–4622, 2018.DOI
- [11]K. Zhang, Y. Li, W. Zuo, L. Zhang, L. V. Gool, and R. Timofte, “Plug-and-play image restoration with deep denoiser prior,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 10, pp. 6360–6376, 2022.arxiv.org/abs/2008.13751
- [12]S. V. Venkatakrishnan, C. A. Bouman, and B. Wohlberg, “Plug-and-play priors for model based reconstruction,” in Proc. IEEE Global Conf. Signal Inf. Process. (GlobalSIP), 2013, pp. 945–948.
- [13]Y. Romano, M. Elad, and P. Milanfar, “The little engine that could: Regularization by denoising (RED),” SIAM J. Imaging Sci., vol. 10, no. 4, pp. 1804–1844, 2017.arxiv.org/abs/1611.02862
- [14]A. Ganesh, Z. Lin, J. Wright, L. Wu, M. Chen, and Y. Ma, “Fast algorithms for recovering a corrupted low-rank matrix,” in Proc. 3rd IEEE Int. Workshop Comput. Adv. Multi-Sensor Adapt. Process. (CAMSAP), 2009, pp. 213–216.DOI
- [15]Z. Lin, M. Chen, and Y. Ma, “The augmented Lagrange multiplier method for exact recovery of corrupted low-rank matrices,” arXiv preprint arXiv:1009.5055, 2010.DOI
- [16]J.-F. Cai, E. J. Candès, and Z. Shen, “A singular value thresholding algorithm for matrix completion,” SIAM J. Optim., vol. 20, no. 4, pp. 1956–1982, 2010.arxiv.org/abs/0810.3286
- [17]T. Bouwmans and E. H. Zahzah, “Robust PCA via principal component pursuit: A review for a comparative evaluation in video surveillance,” Comput. Vis. Image Underst., vol. 122, pp. 22–34, 2014.
- [18]A. Cichocki, D. Mandic, L. De Lathauwer, G. Zhou, Q. Zhao, C. Caiafa, and H. A. Phan, “Tensor decompositions for signal processing applications: From two-way to multiway component analysis,” IEEE Signal Process. Mag., vol. 32, no. 2, pp. 145–163, 2015.DOI
- [19]Y. Zhu, K. Zhang, J. Liang, J. Cao, B. Wen, R. Timofte, and L. V. Gool, “Denoising diffusion models for plug-and-play image restoration,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. Workshops, 2023, pp. 1219–1229.arxiv.org/abs/2305.08995
- [20]X. Fu, K. Huang, W.-K. Ma, N. D. Sidiropoulos, and R. Bro, “Joint tensor factorization and outlying slab suppression with applications,” IEEE Trans. Signal Process., vol. 63, no. 23, pp. 6315–6328, 2015.DOI
- [21]Y. Wang, J. Yang, W. Yin, and Y. Zhang, “A new alternating minimization algorithm for total variation image reconstruction,” SIAM J. Imaging Sci., vol. 1, no. 3, pp. 248–272, 2008.DOI
- [22]K. L. Keys, H. Zhou, and K. Lange, “Proximal distance algorithms: Theory and practice,” Journal of Machine Learning Research, vol. 20, no. 66, pp. 1–38, 2019.pubmed.ncbi.nlm.nih.gov/31649491
- [23]Y. Nesterov, Introductory Lectures on Convex Optimization: A Basic Course. Springer, 2013.
- [24]M. Shao, W.-K. Ma, J. Liu, and Z. Huang, “Accelerated and deep expectation maximization for one-bit MIMO-OFDM detection,” IEEE Trans. Signal Process., vol. 72, pp. 1094–1113, 2024.arxiv.org/abs/2210.03888
- [25]H. Dong, M. Shah, S. Donegan, and Y. Chi, “Deep unfolded tensor robust PCA with self-supervised learning,” in Proc. IEEE Int. Conf. Acoust., Speech Signal Process., 2023, pp. 1–5.arxiv.org/abs/2212.11346
- [26]H. Wang, J. Peng, W. Qin, J. Wang, and D. Meng, “Guaranteed tensor recovery fused low-rankness and smoothness,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 9, pp. 10990–11007, 2023.DOI
- [27]Z. Lai, K. Wei, and Y. Fu, “Deep plug-and-play prior for hyperspectral image restoration,” Neurocomputing, vol. 481, pp. 281–293, 2022.DOI
- [28]Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: From error visibility to structural similarity,” IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, 2004.pubmed.ncbi.nlm.nih.gov/15376593
- [29]R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 586–595.arxiv.org/abs/1801.03924
- [30]C. Peng, Y. Chen, L. Kang, H. Chen, and Q. Cheng, “Hyperspectral image denoising using nonconvex local low-rank and sparse separation with spatial-spectral total variation regularization,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1–17, 2022.DOI