Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PixelFade: Privacy-preserving Person Re-identification with Noise-guided Progressive Replacement

Delong Zhang, Yi-Xing Peng, Xiao-Ming Wu, Ancong Wu, Wei-Shi Zheng

TL;DR

This work tackles privacy leakage in privacy-preserving person re-identification by transforming pedestrian images into noise-like representations that resist recovery attacks while preserving discriminative power under an authorized model. It introduces a Noise-guided Objective Function with a feature-constraint and a heuristic Progressive Pixel Fading strategy that alternates Constraint Operation and Partial Replacement Operation to balance privacy and utility. Empirical results on Market-1501, MSMT17, and CUHK03 show PixelFade outperforms prior PPPR methods in both retrieval accuracy and resistance to recovery, with demonstrated scalability to Text-to-Image and Visible Infrared Re-ID and robustness across backbones. The method provides a practical, adaptable framework for protecting visual privacy in cloud-based Re-ID systems without sacrificing performance for authorized users, supported by extensive ablations and parameter analyses.

Abstract

Online person re-identification services face privacy breaches from potential data leakage and recovery attacks, exposing cloud-stored images to malicious attackers and triggering public concern. The privacy protection of pedestrian images is crucial. Previous privacy-preserving person re-identification methods are unable to resist recovery attacks and compromise accuracy. In this paper, we propose an iterative method (PixelFade) to optimize pedestrian images into noise-like images to resist recovery attacks. We first give an in-depth study of protected images from previous privacy methods, which reveal that the chaos of protected images can disrupt the learning of recovery models. Accordingly, Specifically, we propose Noise-guided Objective Function with the feature constraints of a specific authorization model, optimizing pedestrian images to normal-distributed noise images while preserving their original identity information as per the authorization model. To solve the above non-convex optimization problem, we propose a heuristic optimization algorithm that alternately performs the Constraint Operation and the Partial Replacement Operation. This strategy not only safeguards that original pixels are replaced with noises to protect privacy, but also guides the images towards an improved optimization direction to effectively preserve discriminative features. Extensive experiments demonstrate that our PixelFade outperforms previous methods in resisting recovery attacks and Re-ID performance. The code is available at https://github.com/iSEE-Laboratory/PixelFade.

PixelFade: Privacy-preserving Person Re-identification with Noise-guided Progressive Replacement

TL;DR

This work tackles privacy leakage in privacy-preserving person re-identification by transforming pedestrian images into noise-like representations that resist recovery attacks while preserving discriminative power under an authorized model. It introduces a Noise-guided Objective Function with a feature-constraint and a heuristic Progressive Pixel Fading strategy that alternates Constraint Operation and Partial Replacement Operation to balance privacy and utility. Empirical results on Market-1501, MSMT17, and CUHK03 show PixelFade outperforms prior PPPR methods in both retrieval accuracy and resistance to recovery, with demonstrated scalability to Text-to-Image and Visible Infrared Re-ID and robustness across backbones. The method provides a practical, adaptable framework for protecting visual privacy in cloud-based Re-ID systems without sacrificing performance for authorized users, supported by extensive ablations and parameter analyses.

Abstract

Online person re-identification services face privacy breaches from potential data leakage and recovery attacks, exposing cloud-stored images to malicious attackers and triggering public concern. The privacy protection of pedestrian images is crucial. Previous privacy-preserving person re-identification methods are unable to resist recovery attacks and compromise accuracy. In this paper, we propose an iterative method (PixelFade) to optimize pedestrian images into noise-like images to resist recovery attacks. We first give an in-depth study of protected images from previous privacy methods, which reveal that the chaos of protected images can disrupt the learning of recovery models. Accordingly, Specifically, we propose Noise-guided Objective Function with the feature constraints of a specific authorization model, optimizing pedestrian images to normal-distributed noise images while preserving their original identity information as per the authorization model. To solve the above non-convex optimization problem, we propose a heuristic optimization algorithm that alternately performs the Constraint Operation and the Partial Replacement Operation. This strategy not only safeguards that original pixels are replaced with noises to protect privacy, but also guides the images towards an improved optimization direction to effectively preserve discriminative features. Extensive experiments demonstrate that our PixelFade outperforms previous methods in resisting recovery attacks and Re-ID performance. The code is available at https://github.com/iSEE-Laboratory/PixelFade.
Paper Structure (27 sections, 7 equations, 5 figures, 7 tables, 1 algorithm)

This paper contains 27 sections, 7 equations, 5 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Relative Work
  3. Person Re-Identification
  4. Privacy-preserving Image Recognition
  5. Adversarial Attacks
  6. PixelFade
  7. Noise-guided Objective Function
  8. Progressive Pixel Fading
  9. Constraint Operation
  10. Partial Replacement Operation
  11. Overview and Application of PixelFade
  12. Experiments
  13. Experiments Settings
  14. Datasets.
  15. Threat Models.
  16. ...and 12 more sections

Figures (5)

  • Figure 1: (a) The potential influence of pixel distribution on resisting recovery attacks in protected images. An AD value (from Anderson-Darling razali2011power tests) close to zero signifies that the pixels of the protected image closely align with a normal distribution, signifying more chaos image pixels. Lower SSIM indicates lower quality of the recovered images, signifying stronger resistance to recovery attacks. (b) Visualization of protected and recovered images from different privacy-preserving person re-identification (PPPR) methods.
  • Figure 2: The framework of our PixelFade. Our goal is to optimize the original image $x$ towards noise $\eta$ to obtain the protected image $x^p$. (a) Progressive Pixel Fading alternately run Constraint Operation (CO) and Partial Replacement Operation (PRO) according to the satisfaction of feature constraints. (b) Partial Replacement Operation on the protected images. The randomly generated binary masks $\mathcal{M}^p_t$ are used to select the positions for replacing pixels with noise in the corresponding image.
  • Figure 3: Qualitative results of protected and recovered images from different privacy-preserving PPPR methods. (a) Origin; (b) PrivacyReID zhang2022learnable; (c) Blurring zhang2022learnable; (d) Mosaic zhang2022learnable; (e) AVIH su2023hiding; (f) Our PixelFade.
  • Figure 4: Ablation study of optimization strategy.
  • Figure 5: Parameter analysis of PixelFade. Larger mAP indicates higher Re-ID performance. Smaller SSIM means stronger privacy performance.