NullFace: Training-Free Localized Face Anonymization

2025-03-11

AI Summary

In the digital era, face anonymization in video surveillance faces the fundamental challenge of simultaneously eliminating identity information while preserving non-identity attributes. To address this, we propose a training-free, region-specific face anonymization method leveraging pre-trained text-to-image diffusion models. Our approach employs image inversion and conditional denoising to explicitly edit identity embeddings—enabling fine-grained, user-specified facial region anonymization without model fine-tuning. Crucially, it fully decouples identity from non-identity attributes (e.g., expression, pose, age), ensuring selective and semantically consistent anonymization. Extensive experiments demonstrate that our method surpasses existing state-of-the-art techniques in anonymization strength, attribute fidelity, and visual quality. It exhibits strong robustness across diverse inputs and offers plug-and-play practicality, making it highly suitable for privacy-preserving deployment in real-world surveillance scenarios.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Privacy-enhancing technologies
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Privacy concerns around ever increasing number of cameras are increasing in today's digital age. Although existing anonymization methods are able to obscure identity information, they often struggle to preserve the utility of the images. In this work, we introduce a training-free method for face anonymization that preserves key non-identity-related attributes. Our approach utilizes a pre-trained text-to-image diffusion model without requiring optimization or training. It begins by inverting the input image to recover its initial noise. The noise is then denoised through an identity-conditioned diffusion process, where modified identity embeddings ensure the anonymized face is distinct from the original identity. Our approach also supports localized anonymization, giving users control over which facial regions are anonymized or kept intact. Comprehensive evaluations against state-of-the-art methods show our approach excels in anonymization, attribute preservation, and image quality. Its flexibility, robustness, and practicality make it well-suited for real-world applications. Code and data can be found at https://github.com/hanweikung/nullface .

Problem

• Develops training-free face anonymization preserving non-identity attributes.
• Uses pre-trained diffusion model for localized face anonymization control.
• Ensures high image quality and utility in anonymized facial images.

Innovation

• Training-free face anonymization using diffusion models
• Localized anonymization with user-controlled facial regions
• Identity-conditioned diffusion for distinct anonymized faces