Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis

arXiv:2611.04623 · eess.IV, eess.SP · Submitted 2026-11-03 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: We're looking at the paper: "Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis".

Dev: Here's the gist of the paper. Robust Principal Component Analysis (RPCA) is essential in signal processing, with applications in hyperspectral imaging, face recognition, and video background modeling.

Rosa: This letter proposes a novel Plug-andPlay (PnP) method for multi-channel RPCA that integrates both the physical generative model and cutting-edge deep priors.

Dev: Extensive experiments on RGB and hyperspectral datasets show that PnPPDA 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.

Rosa: Problem Formulation The paper formulates a PnP RPCA formulation with an implicit deep prior, which reads as min X,S γ1Φ(X) + γ2∥X ∥p Sp + γ3∥X ∥TV + λ∥S∥ p 2,p s.t.

Paper summary: Rosa: To recap, this paper, "Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis," is about taking standard RPCA and making it more powerful by plugging in both a physical model and deep priors.

Dev: The core thesis is that traditional RPCA often misses the data priors and struggles with complex image structures, while deep denoisers are good at using data but lack the physical constraints needed for robustness.

Taro: So this paper proposes a novel plug-and-play method, PnP, which integrates both the physical generative model and these cutting-edge deep priors for multi-channel RPCA.

Rosa: They claim that their proposed method, PnPPDA, gets better quantitative performance than conventional RPCA methods and state of the art deep denoisers and tensor approaches while keeping visual details intact.

Dev: The paper formulates a problem where X is the stationary scene or spectral background and S is the small objects or anomalous pixels they want to separate from Y.

Taro: The method uses a novel Plug-and-Play Proximal Distance Algorithm, PnP-PDA, which reformulates it as a consensus optimization problem with subproblems that have closed form updates.

Rosa: They detail how they split the objective function into subproblems for X one, X two, X three and S. For example, the proximal mapping for X one uses a denoiser under the plug-and-play framework.

Dev: The second subproblem for X two is handled by a mode-wise decoupling and averaging strategy with three matrix subproblems.

Taro: The third subproblem for X three involves Fast Total Variation deconvolution, which is used in that part of the algorithm.

Rosa: They show that PnP-PDA achieves better recovery than conventional RPCA methods and state of the art deep denoisers on RGB datasets with consistent quantitative results.

Dev: Plus, they report superior performance on hyperspectral data, showing PnP-PDA gets higher mean peak signal to noise ratio and mean square signed difference across all twelve test conditions.

Taro: Visually, the paper shows that the results preserve textures and edges in the images while removing corruptions.

Conclusion: Rosa: So, looking at "Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis," this method essentially merges a physical model with deep priors to make robust principal component analysis more effective.

Dev: It means that instead of relying only on the data structure or just deep learning priors, you get a solution that leverages both the physical model and those strong deep priors simultaneously.

Taro: The PnPPDA algorithm is what makes this happen by transforming the problem into subproblems for evaluating proximal operators, allowing for closed-form updates and using deep denoiser processing.

Rosa: In simple terms, this means that you get a method that is quantitatively better than previous methods because it uses both the physical understanding of the scene and the learned priors from deep learning models.

Dev: The authors are Ganhua Liu, Mingjie Shao, Wenxuan Sun, Zhi Liu and Yanlong Zhao. They are researchers who focus on this kind of integrated modeling for signal processing applications like hyperspectral imaging or video background modeling.

Taro: The implication is that future work will likely involve testing these methods outside the lab to see how they hold up when the world misbehaves in real-world scenarios and what happens when a system has to recover gracefully from failures.

Guanhua Liu, Mingjie Shao, Wenxuan Sun, Zhi Liu, Yanlong Zhao

School of Information Science and Engineering, Shandong University · State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences

eess.IV, eess.SP

Submitted: 2026-11-03

Updated: 2026-11-03

DOI: 10.1109/LSP.2026.3749560

License: http://creativecommons.org/publicdomain/zero/1.0/

Importance score: 83/100

The gist: Robust Principal Component Analysis (RPCA) is essential in signal processing, with applications in hyperspectral imaging, face recognition, and video background modeling.

Key concepts

Robust Principal Component Analysis (RPCA)
RPCA is a technique used to separate an image or signal into two components: a low-rank component (like the background scene) and a sparse component (like anomalies or noise). It is essential for tasks like denoising and image reconstruction.
Plug-and-Play (PnP) Prior
A PnP prior allows the model to incorporate powerful knowledge from deep learning models, such as denoisers, directly into the mathematical optimization problem. This integrates physical constraints with learned features without needing explicit physical equations for every detail.
Plug-and-Play Proximal Distance Algorithm (PnP-PDA)
This is the core algorithm that solves the RPCA problem by reformulating it as a consensus optimization task. It uses a soft constraint and splits the main problem into four independent subproblems, allowing for efficient solutions using proximal mappings derived from deep priors.

Terminology

Summary

Robust Principal Component Analysis (RPCA) is essential in signal processing, with applications in hyperspectral imaging, face recognition, and video background modeling. p. 0

Where the paper says it

IEEE SIGNAL PROCESSING LETTERS, VOL. X, 2026 1 Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis Guanhua Liu, Mingjie Shao, Wenxuan Sun, Zhi Liu, and Yanlong Zhao Abstract—Robust Principal Component Analysis (RPCA) is essential in signal processing, with applications in hyperspectral imaging, face recognition, and video background modeling.

This letter proposes a novel Plug-andPlay (PnP) method for multi-channel RPCA that integrates both the physical generative model and cutting-edge deep priors. p. 0
Where the paper says it

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-andPlay (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 PnPPDA 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. p. 0
Where the paper says it

This paper proposes a novel Plug-andPlay (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 PnPPDA 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.

Problem Formulation

The paper formulates a PnP RPCA formulation with an implicit deep prior, which reads as min X,S γ1Φ(X) + γ2∥X ∥p Sp + γ3∥X ∥TV + λ∥S∥ p 2,p s.t. Y = X + S. This formulation integrates the physical model with a deep prior Φ(·) through the PnP framework. p. 1

Where the paper says it

Here, xc := vec(X :,:,c) ∈ RMN denotes the vectorization of the c-th spatial channel, and Dh, Dv ∈ RMN×MN 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 Φ(·) through the PnP framework.

The objective is to separate and estimate X and S from Y, where X captures the stationary scene or spectral background and S captures small objects or anomalous pixels. p. 1
Where the paper says it

IEEE SIGNAL PROCESSING LETTERS, VOL. X, 2026 2 X captures the stationary scene or spectral background and S captures small objects or anomalous pixels.

Plug-and-Play Proximal Distance Algorithm

The PnP-PDA algorithm addresses the nonconvex coupling challenge with a novel PDA algorithm that reformulates the task as a consensus optimization problem. The global objective is split into tractable subproblems that allow for efficient closed-form updates. p. 1

Where the paper says it

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.

The algorithm introduces a consensus constraint with X 1, X 2, X 3 = X by C = (X 1, X 2, X 3,S) X 1 = X 2 = X, and then transforms the problem into Fρ(x) = f(x) + ρ squared dist squared (x, C),. p. 1
Where the paper says it

Then, problem (3) is reformulated as min x∈C f(x) = γ1Φ(X 1) + γ2∥X 2∥ p Sp + γ3∥X 3∥TV + λ∥S∥ p 2,p, (4) where x = (X 1, X 2, X 3,S). Next, we transform the consensus constraint with a soft constraint, which is posed as Fρ(x) = f(x) + ρ 2 dist2 (x, C), (5) where dist(x, C) = miny∈C ∥x−y∥F is the distance function and ρ > 0 is a penalty parameter.

This soft constraint is majorized at x t using a quadratic surrogate dist squared (x, C) ≤ x − PC(xt)2F, with equality at x = xt. p. 1
Where the paper says it

Next, we transform the consensus constraint with a soft constraint, which is posed as Fρ(x) = f(x) + ρ 2 dist2 (x, C), (5) where dist(x, C) = miny∈C ∥x−y∥F is the distance function and ρ > 0 is a penalty parameter. We majorize dist2 (x, C) at x t using a quadratic surrogate dist2 (x, C) ≤ ∥x − PC(x t )∥ 2 F , (6) with equality at x = x t , where PC denotes the projection onto C that takes the form PC(x t ) = (P X C , P X C , P X C , P S C ) with P X C = 1 4 X 3 i=1 X t i + Y − S t , P S C = Y − P X C .

Proximal Subproblem Solutions

The PDA update splits into four independent subproblems for X 1, X 2, X 3, and S. For the X 1 subproblem, the proximal mapping is derived as prox(γ1/ρ)Φ(P X C). p. 1

Where the paper says it

1) Proximal Mapping for X 1: The X 1 subproblem reads min X1 γ1Φ(X 1) + ρ 2 ∥X 1 − P X C ∥ 2 F . (9) which amounts to derive the proximal mapping prox(γ1/ρ)Φ(P X C ).

Since Φ is an implicit data prior, its proximal subproblem can be addressed by a denoiser under the PnP framework. p. 1
Where the paper says it

(9) which amounts to derive the proximal mapping prox(γ1/ρ)Φ(P X C ). Since Φ is an implicit data prior, its proximal subproblem can be addressed by a denoiser under the PnP framework [19], as X t+1 1 = Denoiser(P X C , σt), (10) where σt = p γ1/ρ is the noise level parameter.

The X 2 subproblem is approximated using a mode-wise decoupling-and-averaging strategy with three matrix subproblems for k ∈ (1, 2, 3). p. 1
Where the paper says it

2) Proximal Mapping for X 2: The X 2 subproblem is min X2 γ2∥X 2∥ p Sp + ρ 2 ∥X 2 − P X C ∥ 2 F . (11) Since the X2 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 k ∈ {1, 2, 3}, written as min X2(k) γ2∥X2(k)∥ p Sp + ρ 2 ∥X2(k) − P X C(k) ∥ 2 F .

For the X 3 subproblem, the solution involves Fast Total Variation deconvolution (FTVd). p. 2
Where the paper says it

Since the problem is independent across channels, we describe the procedure for each channel. Since Dh and Dv couple the entries of xc, we adopt Fast Total Variation deconvolution (FTVd) [21] by introducing auxiliary variables gh = Dhxc and gv = Dvxc and relaxing the equality constraints with a quadratic penalty of weight β, min xc,gh,gv γ3(∥gh∥1 + ∥gv∥1) + ρ 2 ∥xc − vec(P X C,c)∥ 2 2 + β 2 ∥gh − Dhxc∥ 2 2 + β 2 ∥gv − Dvxc∥ 2 2 .

Performance Comparison

PnP-PDA achieves better recovery performance than conventional RPCA methods and the compared state-of-the-art deep denoisers and tensor RPCA approaches. The PnP-PDA method demonstrates consistently competitive quantitative performance on all ten RGB images. p. 3

Where the paper says it

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.

Furthermore, Table II reports the results on three HSI datasets with 103, 156, and 198 spectral bands, where PnP-PDA achieves the highest MPSNR and MSSIM in all 12 test conditions. p. 3
Where the paper says it

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.

The visual results show that PnPPDA removes corruptions while preserving textures and edges. p. 3
Where the paper says it

Conventional RPCA methods, the ablation variant, and state-ofthe-art tensor RPCA methods produce over-smoothed results, and deep denoisers leave structured artifacts, whereas PnPPDA 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.

Conclusion

This letter presents a PnP method for RPCA that integrates both the physical model and deep priors. Our proposed PnPPDA algorithm transforms the problem into subproblems for evaluating proximal operators, utilizing closed-form updates and deep denoiser processing. p. 3

Where the paper says it

CONCLUSION This letter presents a PnP method for RPCA that integrates both the physical model and deep priors. Our proposed PnPPDA 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. p. 3
Where the paper says it

Our proposed PnPPDA 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.

The PnP-PDA algorithm is summarized in Algorithm 1. p. 2
Where the paper says it

We also adopt a homotopy strategy for ρ by initializing it with a small value and increasing it via ρ t+1 = min(κρt , ρmax) with κ > 1, which gradually enforces the consensus constraints. The complete procedure is summarized in Algorithm 1.

The algorithm complexity analysis is provided in the supplementary material. p. 2
Where the paper says it

The complete procedure is summarized in Algorithm 1. The algorithm complexity analysis is provided in the supplementary material.

REFERENCES

[1] M. M. Salut and D. p. 4

Where the paper says it

X, 2026 5 REFERENCES [1] M. Salut and D.

V. Anderson, “Tensor robust CUR for compression and denoising of hyperspectral data,” IEEE Access, vol. p. 4
Where the paper says it

Salut and D. Anderson, “Tensor robust CUR for compression and denoising of hyperspectral data,” IEEE Access, vol.

11, pp. 77 492–77 505, 2023

[2] J. Wright, A. Y. Yang, A. Ganesh, S. Sastry, and Y Ma, “Robust face recognition via sparse representation,” IEEE Trans. p. 4

Where the paper says it

Sastry, and Y. Ma, “Robust face recognition via sparse representation,” IEEE Trans.

Pattern Anal Mach Intell. p. 4
Where the paper says it

Ma, “Robust face recognition via sparse representation,” IEEE Trans. Pattern Anal.

, vol 31 no 2 pp 210–227, 2009

[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. p. 4

Where the paper says it

Wang, “Tensor robust principal component analysis with side information: Models and applications,” IEEE Trans. Circuits Syst.

, vol 33 no 8 pp 3713–3725, 2023

[4] E. J. Candes, X Li, Y Ma, and J Wright, “Robust principal component analysis? p. 4

Where the paper says it

3713–3725, 2023. Wright, “Robust principal component ` analysis?

” J ACM, vol 58 no 3 pp 1–37, 2011

[5] Z Zhou, X Li, J Wright, E. J Candes, and Y Ma, “Stable principal component pursuit,” in Proc IEEE Int Symp Inf Theory (ISIT), 2010 pp 1518–1522

[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

[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

[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

[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

[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

[12] S V Venkatakrishnan, C A Bouman, and B Wohlberg, “Plug-andplay 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

[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

[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

[16] J.-F Cai, E J Candes, and Z Shen, “A singular value thresholding algorithm for matrix completion,” SIAM J Optim. p. 4

Where the paper says it

Candes, and Z. Shen, “A singular value thresholding ` algorithm for matrix completion,” SIAM J.

, vol 20 no 4 pp 1956–1982, 2010

[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 D 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

[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

[20] X Fu, K Huang, W 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

[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

[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

[23] Y Nesterov, Introductory Lectures on Convex Optimization: A Basic Course. Springer, 2013

[24] M Shao, W 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

[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

[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 10 990–11 007, 2023

[27] Z Lai, K Wei, and Y Fu, “Deep plug-and-play prior for hyperspectral image restoration,” Neurocomputing, vol 481 pp 281–293, 2022

[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

[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

[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.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Deep Plug-and-Play Proximal Distance Algorithm for Robust Principal Component Analysis (PnP-PDA) can be integrated into a multi-channel RPCA formulation by implicitly regularizing it with data priors driven by deep denoisers, specifically using methods like DnCNN or DRUNet. This allows the system to leverage image data priors while maintaining robustness against structured noise, achieving better recovery performance than conventional RPCA methods and state-of-the-art tensor RPCA approaches.

  2. The PnP-PDA algorithm can be used to decouple physical model optimization and prior regularization by reformulating the task as a consensus optimization problem with a soft constraint Frho(x). This allows for the direct embedding of deep denoisers as a modular solver, enabling the system to recover low-rank backgrounds and sparse outliers simultaneously.

  3. The system can improve recovery quality by replacing conventional nuclear norm and l1 norms with nonconvex Schatten-p and l2,p norms for promoting low-rank background and fiber-sparsity structures. This enhancement is further supported by the use of Nesterov extrapolation to accelerate numerical convergence during the optimization process.

  4. For tensor RPCA tasks, the system can utilize a mode-wise decoupling strategy where each mode is solved independently using closed-form reweighted shrinkage updates based on singular values derived from the consensus projection. This allows for efficient recovery of low-rank structures across multiple modes of data simultaneously.

  5. The improved AI system can be tailored to handle hyperspectral image (HSI) datasets, achieving the highest MPSNR and MSSIM in all tested conditions due to its effectiveness with multi-channel constraints and deep priors applied to spectral data.

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.

Sources

Related papers