Multi-challenge database for active liveness

Abstract

Facial authentication on mobile devices has been widely applied in various scenarios. The field of Face Liveness (or Face Anti-Spoofing, FAS) focuses on methods and tools for detecting attacks (spoofs) where a malicious user tries to impersonate someone else or obfuscate their own identity. The specific problem of Active Liveness consists of analyzing the input signal and also the user behavior while performing some required challenge to determine whether the presented face is real or not. Despite the large amount of public datasets in FAS, very few contemplate active liveness, a phenomenon that typically results in new solutions being developed and evaluated with in-house data that cannot be shared due to its sensitive nature. This disjoint character in evaluations leads to irreproducible works and greatly hinders the mutual contribution inside the scientific community. In this paper, we describe an approach for creating a new database for active liveness detection. In this database, volunteers use a mobile application (under alpha testing for Android and iOS devices) to record themselves executing three distinct interactions (challenges) with their heads, namely face close up, head orientation, and flashes. Users will be encouraged to acquire videos in different environments and times of day, which will contribute to high variance in the dataset. After all acquisitions, we will split the dataset into training and testing protocols.