“…Due to the increasing number of pub-150 lic benchmark databases, numerous passive software-based approaches have been proposed for face anti-spoofing. In general, passive methods based on analyzing different facial properties, like frequency content [33,34], texture [10,35,36,37,38,39] and quality [40,41,42], or motion cues, like eye blinking 155 [43,44,45,46], facial expression changes [27,44,45,46], mouth movements [27,44,45,46], or even color variation due to blood circulation (pulse) [47], to discriminate face artifacts from genuine ones. Passive software-based methods have shown impressive results on the publicly available datasets but 160 the preliminary cross-database tests, like [11,32], revealed that the performance is likely to degrade drastically when operating in unknown conditions.…”