Provably Contractive and High-Quality Denoisers for Convergent Restoration

2026-03-27

AI Summary

This work proposes the first provably contractive image denoising network with a strict global 1-Lipschitz constraint, addressing the instability of existing denoisers under input perturbations and the trade-off between robustness and reconstruction fidelity. By unrolling proximal optimization steps, the method introduces Lipschitz-controlled convolutional refinement modules that ensure mapping contractiveness while enabling high-quality image recovery. The resulting model achieves state-of-the-art performance among 1-Lipschitz-constrained architectures, matching the accuracy of unconstrained state-of-the-art approaches, while significantly enhancing stability against input perturbations. Furthermore, the framework provides theoretical convergence guarantees for Plug-and-Play algorithms, bridging practical performance with formal robustness assurances.

Abstract

Image restoration, the recovery of clean images from degraded measurements, has applications in various domains like surveillance, defense, and medical imaging. Despite achieving state-of-the-art (SOTA) restoration performance, existing convolutional and attention-based networks lack stability guarantees under minor shifts in input, exposing a robustness accuracy trade-off. We develop provably contractive (global Lipschitz $< 1$) denoiser networks that considerably reduce this gap. Our design composes proximal layers obtained from unfolding techniques, with Lipschitz-controlled convolutional refinements. By contractivity, our denoiser guarantees that input perturbations of strength $\|δ\|\le\varepsilon$ induce at most $\varepsilon$ change at the output, while strong baselines such as DnCNN and Restormer can exhibit larger deviations under the same perturbations. On image denoising, the proposed model is competitive with unconstrained SOTA denoisers, reporting the tightest gap for a provably 1-Lipschitz model and establishing that such gaps are indeed achievable by contractive denoisers. Moreover, the proposed denoisers act as strong regularizers for image restoration that provably effect convergence in Plug-and-Play algorithms. Our results show that enforcing strict Lipschitz control does not inherently degrade output quality, challenging a common assumption in the literature and moving the field toward verifiable and stable vision models. Codes and pretrained models are available at https://github.com/SHUBHI1553/Contractive-Denoisers

Problem

• image restoration
• robustness-accuracy trade-off
• input perturbations
• stability guarantees
• denoiser networks

Innovation

• contractive denoiser
• Lipschitz-constrained networks
• provably stable restoration
• Plug-and-Play convergence
• robust image denoising