GANonymization: A GAN-based Face Anonymization Framework for Preserving Emotional Expressions

TL;DR

GANonymization introduces a landmark-based intermediate representation (478 3D landmarks) and a pix2pix-based re-synthesis to anonymize faces while preserving emotional expressions. By discarding nonessential contextual information before generation and avoiding noise-based differential privacy, it achieves high-quality, anonymized faces that retain emotion-related cues better than several baselines in many cases. The approach also analyzes which facial traits are removed (e.g., hair, jewelry) versus preserved (e.g., Smiling), revealing bias-reduction benefits and data-distribution influences. Overall, the method offers a practical privacy-preserving pipeline for emotion-centric applications with strong face naturalness and targeted trait removal, suggesting applicability beyond emotion tasks and potential extensions to healthcare and other domains.

Abstract

In recent years, the increasing availability of personal data has raised concerns regarding privacy and security. One of the critical processes to address these concerns is data anonymization, which aims to protect individual privacy and prevent the release of sensitive information. This research focuses on the importance of face anonymization. Therefore, we introduce GANonymization, a novel face anonymization framework with facial expression-preserving abilities. Our approach is based on a high-level representation of a face, which is synthesized into an anonymized version based on a generative adversarial network (GAN). The effectiveness of the approach was assessed by evaluating its performance in removing identifiable facial attributes to increase the anonymity of the given individual face. Additionally, the performance of preserving facial expressions was evaluated on several affect recognition datasets and outperformed the state-of-the-art methods in most categories. Finally, our approach was analyzed for its ability to remove various facial traits, such as jewelry, hair color, and multiple others. Here, it demonstrated reliable performance in removing these attributes. Our results suggest that GANonymization is a promising approach for anonymizing faces while preserving facial expressions.

Figures (11)

• Figure 1: Existing privacy preservation concepts in the context of face anonymization.
• Figure 2: Architecture of the GANonymization pipeline.
• Figure 3: Sample of synthesized faces based on the WIDER dataset.
• Figure 4: Sample of synthesized faces based on the AffectNet dataset.
• Figure 5: Sample of synthesized faces based on the CK+ dataset.