Master Face Attacks on Face Recognition Systems

Abstract

Due to its simplicity, face authentication is currently more frequently used than authentication using a personal identification number or an unlock pattern, especially on mobile devices. This has made it a seductive target for attackers who use a demonstration assault. Traditional presentation attacks employ the victim's face or victim footage. The existence of master faces—faces that match numerous enrolled templates in face recognition systems—has been demonstrated in earlier research, and their presence increases the effectiveness of presentation attacks. In this article, we present the results of a thorough investigation of latent variable evolution (LVE), a technique frequently employed to produce master faces. To determine the characteristics of master faces, an LVE algorithm was used in a variety of settings and with many databases and/or face recognition systems.