Transient Authentication from First-Person-View Video

We propose PassFrame, a system which utilizes first-person-view videos to generate personalized authentication challenges based on human episodic memory of event sequences. From the recorded videos, relevant (memorable) scenes are selected to form image-based authentication challenges. These authentication challenges are compatible with a variety of screen sizes and input modalities. As the popularity of using wearable cameras in daily life is increasing, PassFrame may serve as a convenient personalized authentication mechanism to screen-based appliances and services of a camera wearer. We evaluated the system in various settings including a spatially constrained scenario with 12 participants and a deployment on smartphones with 16 participants and more than 9 hours continuous video per participant. The authentication challenge completion time ranged from 2.1 to 9.7 seconds (average: 6 sec), which could facilitate a secure yet usable configuration of three consecutive challenges for each login. We investigated different versions of the challenges to obfuscate potential privacy leakage or ethical concerns with 27 participants. We also assessed the authentication schemes in the presence of informed adversaries, such as friends, colleagues or spouses and were able to detect attacks from diverging login behaviour.