Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NFCDS: A Plug-and-Play Noise Frequency-Controlled Diffusion Sampling Strategy for Image Restoration

Zhen Wang, Hongyi Liu, Jianing Li, Zhihui Wei

TL;DR

The paper analyzes how injected noise in diffusion sampling affects fidelity and perceptual quality for zero-shot image restoration, revealing that low-frequency noise harms global fidelity while high-frequency noise enables texture. It proposes NFCDS, a Fourier-domain masking mechanism that suppresses low-frequency components of the injected noise after data-consistency steps, preserving high-frequency content and injecting a data-consistency prior without retraining. NFCDS is training-free and plug-and-play, demonstrating improved reconstruction fidelity (PSNR/SSIM) and competitive perceptual quality (LPIPS) across super-resolution and denoising tasks, with faster convergence and reduced sampling steps. The approach generalizes across diffusion-based PnP restoration frameworks, offering a practical and scalable solution for high-fidelity, perceptually convincing zero-shot image restoration.

Abstract

Diffusion sampling-based Plug-and-Play (PnP) methods produce images with high perceptual quality but often suffer from reduced data fidelity, primarily due to the noise introduced during reverse diffusion. To address this trade-off, we propose Noise Frequency-Controlled Diffusion Sampling (NFCDS), a spectral modulation mechanism for reverse diffusion noise. We show that the fidelity-perception conflict can be fundamentally understood through noise frequency: low-frequency components induce blur and degrade fidelity, while high-frequency components drive detail generation. Based on this insight, we design a Fourier-domain filter that progressively suppresses low-frequency noise and preserves high-frequency content. This controlled refinement injects a data-consistency prior directly into sampling, enabling fast convergence to results that are both high-fidelity and perceptually convincing--without additional training. As a PnP module, NFCDS seamlessly integrates into existing diffusion-based restoration frameworks and improves the fidelity-perception balance across diverse zero-shot tasks.

NFCDS: A Plug-and-Play Noise Frequency-Controlled Diffusion Sampling Strategy for Image Restoration

TL;DR

The paper analyzes how injected noise in diffusion sampling affects fidelity and perceptual quality for zero-shot image restoration, revealing that low-frequency noise harms global fidelity while high-frequency noise enables texture. It proposes NFCDS, a Fourier-domain masking mechanism that suppresses low-frequency components of the injected noise after data-consistency steps, preserving high-frequency content and injecting a data-consistency prior without retraining. NFCDS is training-free and plug-and-play, demonstrating improved reconstruction fidelity (PSNR/SSIM) and competitive perceptual quality (LPIPS) across super-resolution and denoising tasks, with faster convergence and reduced sampling steps. The approach generalizes across diffusion-based PnP restoration frameworks, offering a practical and scalable solution for high-fidelity, perceptually convincing zero-shot image restoration.

Abstract

Diffusion sampling-based Plug-and-Play (PnP) methods produce images with high perceptual quality but often suffer from reduced data fidelity, primarily due to the noise introduced during reverse diffusion. To address this trade-off, we propose Noise Frequency-Controlled Diffusion Sampling (NFCDS), a spectral modulation mechanism for reverse diffusion noise. We show that the fidelity-perception conflict can be fundamentally understood through noise frequency: low-frequency components induce blur and degrade fidelity, while high-frequency components drive detail generation. Based on this insight, we design a Fourier-domain filter that progressively suppresses low-frequency noise and preserves high-frequency content. This controlled refinement injects a data-consistency prior directly into sampling, enabling fast convergence to results that are both high-fidelity and perceptually convincing--without additional training. As a PnP module, NFCDS seamlessly integrates into existing diffusion-based restoration frameworks and improves the fidelity-perception balance across diverse zero-shot tasks.
Paper Structure (10 sections, 17 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 10 sections, 17 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Denoising Diffusion Probabilistic Model
  4. Restoration via Plug-and-Play Diffusion Models
  5. Method
  6. Role of Injected Noise in Image Generation
  7. Role of Injected Noise in Image Restoration
  8. Noise Frequency-Controlled Diffusion Sampling
  9. Experiment
  10. Conclusion

Figures (4)

  • Figure 1: Role of Low-Frequency Noise in Generation.
  • Figure 2: Role of High-Frequency Noise in Generation.
  • Figure 3: Intermediate processes of generation and restoration.
  • Figure 4: Visual comparison of results (N/A: method is unable to handle the task).