Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Key-Driven Framework for Identity-Preserving Face Anonymization

Miaomiao Wang, Guang Hua, Sheng Li, Guorui Feng

TL;DR

The paper tackles the privacy-utility conflict in virtual face content by introducing a key-driven framework (KFAAR) that simultaneously anonymizes and authenticates faces. It couples a head posture-preserving generator (HPVFG) with a key-controlled authenticator (KVFA) trained via multitask objectives to ensure anonymity, synchronism, and diversity while enabling identity retrieval with the correct key. Experimental results on LFW, CelebA, and FFHQ show strong anonymization, high synchronism, realistic visuals, and robust authentication performance, along with an in-depth security analysis of key-based risks and edge-case handling. The approach provides practical, interactive, and secure identity management for metaverse face content, offering a pathway to privacy-preserving yet verifiable virtual identities.

Abstract

Virtual faces are crucial content in the metaverse. Recently, attempts have been made to generate virtual faces for privacy protection. Nevertheless, these virtual faces either permanently remove the identifiable information or map the original identity into a virtual one, which loses the original identity forever. In this study, we first attempt to address the conflict between privacy and identifiability in virtual faces, where a key-driven face anonymization and authentication recognition (KFAAR) framework is proposed. Concretely, the KFAAR framework consists of a head posture-preserving virtual face generation (HPVFG) module and a key-controllable virtual face authentication (KVFA) module. The HPVFG module uses a user key to project the latent vector of the original face into a virtual one. Then it maps the virtual vectors to obtain an extended encoding, based on which the virtual face is generated. By simultaneously adding a head posture and facial expression correction module, the virtual face has the same head posture and facial expression as the original face. During the authentication, we propose a KVFA module to directly recognize the virtual faces using the correct user key, which can obtain the original identity without exposing the original face image. We also propose a multi-task learning objective to train HPVFG and KVFA. Extensive experiments demonstrate the advantages of the proposed HPVFG and KVFA modules, which effectively achieve both facial anonymity and identifiability.

A Key-Driven Framework for Identity-Preserving Face Anonymization

TL;DR

The paper tackles the privacy-utility conflict in virtual face content by introducing a key-driven framework (KFAAR) that simultaneously anonymizes and authenticates faces. It couples a head posture-preserving generator (HPVFG) with a key-controlled authenticator (KVFA) trained via multitask objectives to ensure anonymity, synchronism, and diversity while enabling identity retrieval with the correct key. Experimental results on LFW, CelebA, and FFHQ show strong anonymization, high synchronism, realistic visuals, and robust authentication performance, along with an in-depth security analysis of key-based risks and edge-case handling. The approach provides practical, interactive, and secure identity management for metaverse face content, offering a pathway to privacy-preserving yet verifiable virtual identities.

Abstract

Virtual faces are crucial content in the metaverse. Recently, attempts have been made to generate virtual faces for privacy protection. Nevertheless, these virtual faces either permanently remove the identifiable information or map the original identity into a virtual one, which loses the original identity forever. In this study, we first attempt to address the conflict between privacy and identifiability in virtual faces, where a key-driven face anonymization and authentication recognition (KFAAR) framework is proposed. Concretely, the KFAAR framework consists of a head posture-preserving virtual face generation (HPVFG) module and a key-controllable virtual face authentication (KVFA) module. The HPVFG module uses a user key to project the latent vector of the original face into a virtual one. Then it maps the virtual vectors to obtain an extended encoding, based on which the virtual face is generated. By simultaneously adding a head posture and facial expression correction module, the virtual face has the same head posture and facial expression as the original face. During the authentication, we propose a KVFA module to directly recognize the virtual faces using the correct user key, which can obtain the original identity without exposing the original face image. We also propose a multi-task learning objective to train HPVFG and KVFA. Extensive experiments demonstrate the advantages of the proposed HPVFG and KVFA modules, which effectively achieve both facial anonymity and identifiability.
Paper Structure (49 sections, 2 equations, 10 figures, 12 tables)

This paper contains 49 sections, 2 equations, 10 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Our Work
  4. Related Work
  5. Problem Statement
  6. System Model
  7. The Interaction Process
  8. Threat Model
  9. Key Guessing
  10. Model Guessing
  11. The Proposed Method
  12. Overview
  13. Virtual Face Generation
  14. Properties of Virtual Faces
  15. Network Architecture of HPVFG
  16. ...and 34 more sections

Figures (10)

  • Figure 1: Illustration of the proposed framework. The virtual face generated by our HPVFG has significant differences in terms of visual appearance and machine recognition from the original face. With the correct key, it can be authenticated through KVFA to match their original identity.
  • Figure 2: System Model. The participants are user (U), face anonymization server (FAS), face recognizer (FR), virtual face authentication server (VFAS) and adversary (AD).
  • Figure 3: The proposed KFAAR framework for virtual face generation and authentication. (a) HPVFG for virtual face generation, (b) KVFA for virtual face authentication.
  • Figure 4: Illustration of our training strategy. (a) Virtual Face Generation, (b) virtual Face Authentication.
  • Figure 5: Examples of the virtual faces. The first two rows are the original face images, while the third and fourth rows are the virtual face images of the first and second rows, respectively.
  • ...and 5 more figures