Faces at Face Value: An Analysis of Face Recognition Technology Policy and Performance

Abstract

Facial recognition is one of the best developed and widely used applications of machine learning and examples of artificial intelligence in 2023. What was once a distant idea of futurism is now used in nearly every smart phone for identity verification. Alongside the convenient uses of the technology stand its more Orwellian counterparts - most notably, the use of face recognition for public surveillance. The development of the technology and the ubiquity of high-quality video recording devices like traffic cameras, surveillance cameras, and police body cameras enables the permeation of this technology throughout all spheres of life. Such technology invites the fear of constant surveillance and the decline in individual privacy, particularly in public areas. While the field has been subject to significant research and policy interest, it remains insufficiently regulated and misunderstood from a technological perspective. This study aims to comprehensively gauge the current policies surrounding the use of face recognition - specifically regarding its implementation on police body cameras and public footage as well as the use of personal photos in face recognition databases. Additionally, this study aims to quantify its accuracy in the face of adversarial factors - specifically, relating to age-invariant cross-demographic identity tracking i.e. identity matching with photos from different ages across different ethnic groups.