Seeing is not Believing: An Identity Hider for Human Vision Privacy Protection

TL;DR

This work tackles privacy concerns arising when large facial image databases are reviewed by humans, by proposing an identity hider that simultaneously conceals appearance from human observers and preserves identifiability for machine vision. It introduces a two-module framework: a Virtual Face Generation Module that uses StyleGAN2 latent space $\\mathcal{Z}^+$ to create a virtual face with a new appearance while maintaining a similar parsing map, and an Appearance Transfer Module (DisenNet) that transfers this appearance into the original face via identity and attribute disentanglement losses. The model is trained with a combination of adversarial, identity-disentanglement, attribute-disentanglement, visual-content, and reconstruction losses, collectively optimizing $L_{total}$; it also supports diversity via style-mixing and background-preservation through a parsing-based masking strategy. Experimental results on CelebA-HQ and VGGFace2 show that the proposed identity hider achieves stronger human-vision privacy protection and transferable identifiability for machine recognizers, outperforming PRO-Face on several metrics and enabling robust parsing-map preservation and background handling, which broadens practical applicability.

Abstract

Massive captured face images are stored in the database for the identification of individuals. However, these images can be observed unintentionally by data managers, which is not at the will of individuals and may cause privacy violations. Existing protection schemes can maintain identifiability but slightly change the facial appearance, rendering it still susceptible to the visual perception of the original identity by data managers. In this paper, we propose an effective identity hider for human vision protection, which can significantly change appearance to visually hide identity while allowing identification for face recognizers. Concretely, the identity hider benefits from two specially designed modules: 1) The virtual face generation module generates a virtual face with a new appearance by manipulating the latent space of StyleGAN2. In particular, the virtual face has a similar parsing map to the original face, supporting other vision tasks such as head pose detection. 2) The appearance transfer module transfers the appearance of the virtual face into the original face via attribute replacement. Meanwhile, identity information can be preserved well with the help of the disentanglement networks. In addition, diversity and background preservation are supported to meet the various requirements. Extensive experiments demonstrate that the proposed identity hider achieves excellent performance on privacy protection and identifiability preservation.

Figures (14)

• Figure 1: A protection example of the proposed identity hider. For the protected facial images of Bob, they have similar parsing maps but new appearances, rendering it difficult for data examiners to identify them as Bob via human vision, but easy for face recognizers via machine vision.
• Figure 2: Visual samples of our hider and mainstream schemes. Below the pair of original and protected faces shows their identity similarities via face comparing of Face++, where the matching threshold is 74.0%.
• Figure 3: A face recognition system.
• Figure 4: The process of the identity hider, which contains the virtual face generation module (VFGM) and the appearance transfer module (ATM). VFGM generates a virtual face with a new visual appearance and the similar parsing map with the original face. ATM transfers the visual appearance of the virtual face into the original face via the disentanglement networks.
• Figure 5: Style-mixing for diverse results.