Facial RECognition Prevention

Management of Security

Beyond the national defense, the army is the mirror image of the country, its defense, and as such, professionalism, credibility, and reputation are of utmost importance. A public show of force are official ceremonies, parades, and the celebration of the national day, and many other events. These appearances symbolize the nation’s strength, resilience, and commitment to its values.

During public events such as parades and ceremonies, army members are often photographed. The photos are then published in newspapers, magazines, and on social media. Although this practice is good for publicity, it creates a risk for the members of the army and the army’s activities in general.

How members of the army personnel can be identified? Military members can be present on social media platforms (with their own private profile or through profiles of family members or friends). With new advances in artificial intelligence, an adversary could identify a large number of military personnel in an automated fashion. An adversary could take official pictures of army members that are publicly shared and match them with social media profiles to retrieve additional information (e.g., name, address, friends, and family details).

If malicious actors retrieve such private information, the army personnel can face several apparent risks, such as:

1. Personal Security Risks: Military members and their families are vulnerable to targeted threats, harassment, blackmailing, or even physical harm from malicious actors;
2. Cybersecurity Threats: Spear phishing, social engineering, and other cyberattacks may be directed at individuals through their social media profiles
3. Influence Operations: Adversarial nations or groups may attempt to influence or manipulate military members through social media, spreading misinformation, propaganda, or extremist ideologies.

The primary objective of this project is to devise new adversarial AI methods to make the facial recognition of members of the army more difficult. To that end, the following research directions will be explored:

• Apparel accessories recommendation to fool AI-based models: a first direction to explore is to propose and assess the impact of using accessories such as scarves, glasses, caps, or makeup to make the use of AI facial recognition techniques harder.
• Automated alert system via AI-based monitoring of social media: A second direction to explore is to provide advice and recommendations to the army and its members regarding their presence on social media. The goal is to develop an automated inspection tool for social media to detect pictures that could allow the identification of the members of the army.
• Invisible “blurring” of pictures: A third research direction is more exploratory. The goal is to develop a tool that could be shared with army members (and their families) for systematically post-processing their private photos before sharing them online. The idea of the post-processing is to modify (e.g. a sort of “blurring”) them in a way that makes automated face recognition with AI very challenging.

Publications

• Alexandre De Groodt. 2025. Facial Recognition Prevention: Face web scraping. Université Libre de Bruxelles
• Pawel Borsukiewicz, Fadi Boutros, Iyiola Olatunji, Charles Beumier, Wendkûuni Ouédraogo, Jacques Klein and Tegawendé Bissyandé. 2026. Beyond real faces: synthetic datasets can achieve reliable recognition performance without privacy compromise. NPJ Artificial Intelligence, 2
• Pawel Borsukiewicz, El-hacen Diallo, Abdoul Aziz Bonkoungoun, Wendkûuni Ouédraogo, Jacques Klein, Charles Beumier and Tegawendé Bissyandé. 2026. Faces in Focus: Evaluating Generalizability and Transferability of Facial Recognition Explainability Techniques for CNN-based Models. Proceedings of the IEEE/CVF Conference on CVPR Workshops 2026