Year: 2025
Type: article
Abstract: Driven by deep learning, face recognition technology has become increasingly advanced and is now applied in various fields. However, this technology still faces challenges regarding fairness and privacy protection. Due to multiple factors, such as the imbalance of training data, there are significant discrepancies in the recognition accuracy of facial recognition systems across different populations, leading to legal issues. Privacy risks such as unauthorized data collection and identity theft have garnered widespread attention. This paper systematically analyzes the bias present in current mainstream face recognition models and discusses fairness optimization methods, such as data balancing, feature learning improvements, and multi-task learning. Simultaneously, the paper reviews the latest privacy protection technologies to mitigate data security risks, including federated learning, differential privacy, and homomorphic encryption. By exploring fairness and privacy-preserving technologies, this study aims to promote the development of face recognition technology toward a more equitable and secure direction.
Author: Jiahui Lin
Institution: Xiamen University Malaysia
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: hybrid
OpenAlex ID: W4409237114