Abstract

Face  anti-spoofing  is  essential  to  prevent  face  recog-nition systems from a security breach.   Much of the pro-gresses  have  been  made  by  the  availability  of  face  anti-spoofing  benchmark  datasets  in  recent  years.   However,

existing face anti-spoofing benchmarks have limited num-ber  of  subjects  (≤170)  and  modalities  (≤2),  whichhinder  the  further  development  of  the  academic  commu-nity.   To  facilitate  face  anti-spoofing  research,  we  intro-duce  a  large-scale  multi-modal  dataset,  namely  CASIA-SURF, which is the largest publicly available dataset for face  anti-spoofing  in  terms  of  both  subjects  and  visual modalities.  Specifically, it consists of 1,000subjects with21,000videos and each sample has

3modalities (i.e., RGB,Depth and IR). We also provide a measurement set, evalu-ation protocol and training/validation/testing subsets, de-veloping a new benchmark for face anti-spoofing.  More-over, we present a new multi-modal fusion method as base-line,  which  performs  feature  re-weighting  to  select  the more  informative  channel  features  while  suppressing  the

less  useful  ones  for  each  modal.   Extensive  experiments have  been  conducted  on  the  proposed  dataset  to  verify

its significance and generalization capability.  The dataset is  available  at https://sites.google.com/qq.com/chalearnfacespoofingattackdete/