Detection of Face Spoofing Using Visual Dynamics

摘要: Rendering a face recognition system robust is vital in order to safeguard it against spoof attacks carried out using printed pictures of victim (also known as print attack) or replayed video the person (replay attack). A key property distinguishing live, valid access from media videos by exploiting information dynamics content, such blinking eyes, moving lips, and facial dynamics. We advance state art antispoofing applying recently developed algorithm called dynamic mode decomposition (DMD) general purpose, entirely data-driven approach capture above liveness cues. propose classification pipeline consisting DMD, local binary patterns (LBPs), support vector machines (SVMs) with histogram intersection kernel. unique DMD its ability conveniently represent temporal entire single image same dimensions those images contained video. The + LBP SVM proves be efficient, convenient use, effective. In fact only spatial configuration for needs tuned. effectiveness methodology was demonstrated three publicly available databases: 1) print-attack; 2) replay-attack; 3) CASIA-FASD, attaining comparable results art, following respective published experimental protocols.