3D mask face anti-spoofing with remote photoplethysmography

A method for a 3-dimensional mask face anti-spoofing of a subject of interest comprising:

• capturing a video of a face of the subject of interest with an RGB camera configured to capture blood color variation when light is reflected or absorbed by semi-transparent human skin;
• measuring human blood flow variation caused by a heartbeat by extracting local remote photoplethysmography signals from the captured video of the face of the subject of interest to produce extracted local remote photoplethysmography signals;
• extracting the extracted local remote photoplethysmography signals through cross-correlation to produce a robust cross-correlation of the extracted local remote photoplethysmography signals for 3-dimensional mask face anti-spoofing through a union of similarity of all possible signals combination defined as
• x=Ui,j=1, . . . N,i≤jρ(si,sj)
• where ρ(si, s1) is a similarity between two signals si and sj, and union U is a concatenation operator, and the similarity ρ(si,sj) is the maximum value of cross-correlation spectrum of two local remote photoplethysmography signals, and wherein
• ρ(si,sj)=max|{si★sj}|
• where F is Fourier transform and * is a cross-correlation operator;
• learning a confidence map based on the extracted local remote photoplethysmography signals;
• and
• classifying the captured video of the face of the subject based on the cross-correlation of the extracted local remote photoplethysmography signals and the confidence map to determine if a 3-dimensional mask face is used.