Towards federated learning for face recognition: a comprehensive review of secure and privacy-preserving techniques and prospective research

Abstract

Federated learning is a popular decentralized machine learning paradigm that enables multiple devices or entities to collectively train a common model without sending data to a central server. The distributed data availability improves privacy and security for each participating entity by keeping their data localized. Face recognition, a rapidly evolving field, uses deep learning techniques to achieve remarkable advances, frequently outperforming human performance on specific public datasets. However, the sensitive nature of facial data, as faces belong to personally identifiable information, raises serious privacy and security concerns on training data sharing. Federated learning has emerged as a promising solution to the privacy and security issues confronting the face recognition community. This paper provides a systematic, PRISMA-guided review of 50 publications released between 2020 and 2026, identified through searches of IEEE Xplore, ACM Digital Library, SpringerLink, arXiv, Google Scholar, and DBLP, on face recognition frameworks that use federated learning. The reviewed techniques are systematically classified into six categories, with an emphasis on their contributions and relevance to the larger domain of federated learning based face recognition. This work aims to summarize and analyze various federated facial recognition methods, including their underlying techniques, results, and goals. This survey also compares against twelve existing face recognition surveys to clarify its distinct and unique contributions. Furthermore, it provides a high-level detailed overview of how various federated learning functionalities and design principles have been used in face recognition applications. This review identifies key challenges while highlighting promising research directions for future advancements in the field.