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

Fabio Hellmann,Silvan Mertes,Mohamed Benouis,Alexander Hustinx,Tzung-Chien Hsieh,Cristina Conati,Peter Krawitz,Elisabeth André

DOI: https://doi.org/10.1145/3641107

2023-11-14

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.

Computer Vision and Pattern Recognition,Multimedia

What problem does this paper attempt to address?

The paper aims to address the issue of preserving emotional expressions during facial anonymization. Specifically:

- **Privacy Protection**: With the increasing amount of personal data, privacy and security issues have garnered widespread attention. Data anonymization is a key process aimed at protecting personal privacy and preventing the leakage of sensitive information. However, when dealing with facial data, existing anonymization techniques often fail to effectively remove features that can be used to re-identify individuals (such as skin color, gender, or age).

- **Anonymization in Emotion Recognition Tasks**: The paper proposes a new method based on Generative Adversarial Networks (GAN) — GANonymization, which aims to remove facial features that can be used to re-identify individuals while retaining important information for emotion recognition. Unlike existing methods, this approach extracts facial key points as an intermediate representation and uses GAN to re-synthesize facial images, thereby removing irrelevant features such as hairstyle, skin color, and gender.

- **Ethical and Legal Compliance**: Facial anonymization not only helps protect personal privacy but also promotes ethical considerations and ensures compliance with relevant laws and regulations. For example, in medical records and criminal justice data, anonymization is crucial for protecting personal privacy. Additionally, the European Union's General Data Protection Regulation (GDPR) mandates the protection of personal information through anonymization or pseudonymization.

Through the above measures, researchers can prevent the misuse of personal information and develop more widely applicable and ethically compliant machine learning models.