An Evolutionary Face Recognition and Verification System 16.03


Project Start Date: Jul 1, 2016
Funding: Member Funded
Project Tags: ,

Project Summary

The  use  of  face  verification  systems  as  a  primary  source  of  authentication  has  been  very  common  over  past  few  years.  Despite  the  advance  in   face  recognition  systems,  there  are  still  many  open  breaches  left  in  this  domain.  One  of  the  practical  challenges  is  to  secure  face  biometric  systems  from   intruder’s  attacks,  where  an  unauthorized  person  tries  to  gain  access  by  showing  the  counterfeit  evidence  (images/videos)  in  front  of  face  biometric  system.   We  propose  a  novel  approach,  which  can  be  easily  integrated  to  the  existing  face  verification  systems  without  any  additional  hardware  deployment.  Systems   that  deliver  the  power  to  authenticate  persons  accurately,  swiftly,  reliably,  without  invading  privacy  issues,  cost  effectively,  in  a  user-­‐friendly  manner  and   without  requiring  radical  modifications  to  the  existing  infrastructures  are  desired.  This  field  of  detection  of  imposter  attempts  is  an  open  research  problem,  as   more  sophisticated  and  advanced  spoofing  attempts  come  into  play.  Current  anti-­‐spoofing  methods  suffer  various  problems  that  make  them  unreliable  and   inadequate  to  integrate  them  with  face  recognition  systems.  We  shall  attack  this  problem  within  two  research  scenarios  over  the  distinct  classification   schemes:  Learning  with  a  large  but  a  single  Operational  Neural  Network  (ONN)  and  ‘Divide  and  Conquer’  Learning  with  ensembles  of  simple  but  ‘expert’   ONNs,  each  of  which  dedicated  for  a  subset  of  the  dataset  with  a  certain  homogeneity.