Interactive Photo Liveness for Presentation Attacks Detection

This paper presents an interactive liveness detection approach against presentation attacks. It aims to minimize the impact on the user, who is only asked to produce single head movements. The described approach combines two methods: (1) single-photo liveness estimation based on CNN implementation, and (2) interactive liveness estimation based on head movements detected from two video frames extracted before and during the movement. The resulting system is designed to work on smartphones and by web-cameras. An appropriate database was collected for experiments. These achieved EER of less than 5% for paper spoofing attacks, less than 4% for monitor and 0.6% for tablet, while the Failure to Capture (FTC) was less than 3% for the most user-friendly scenario.