Susang Kim(healess1@gmail.com)
Anti-Spoofing
Learning Deep Models for Face Anti-Spoofing:
Binary or Auxiliary Supervision
Anti-Spoofing 관련 아래 논문을 바탕으로 설명
[논문]
Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision
In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2018), Salt Lake City, UT, Jun. 2018
http://cvlab.cse.msu.edu/pdfs/Liu_Jourabloo_Liu_CVPR2018.pdf
[Data]
Spoof in the Wild (SiW) Face Anti-spoofing Database
http://cvlab.cse.msu.edu/spoof-in-the-wild-siw-face-anti-spoofing-database.html
얼굴인식 실제 서비스로 이어지기 위해서는...
실재 얼굴과 Print,Phone(Galaxy,iPhone),
가면, 동영상등을 구분해야함
[출처] : http://biometrics.cse.msu.edu/projects/face_recog_spoofing.html
By Auxiliary Supervision
RGB이미지를 바탕으로 Depth정보를 생성하고 rPPG정보로 분류
What is rPPG?
[출처] : https://www.noldus.com/facereader/remote-photoplethysmography-facereader
Typical Application of rPPG
Desktop implementation of Remote Photoplethysmography – Measuring heart rate using facial video.
(https://github.com/prouast/heartbeat)
Proposed Method
CNN으로 이미지 정보를 학습 rPPG정보를 RNN으로 학습하여 구성
Network Architecture (CNN+RNN)
Depthmap (3D Reconstruction)
rPPG Supervision
Non-rigid Registration
Feature Map과 DepthMap 그리고 3D Shape의 3가지 Input
=> frontalized F output
Spoof in the Wild Database
data from live and spoof viedos
Anti-Spoof Data
Datasets # of subj. / # of
sess.
Links Year Spoof attacks
attacks
Publish
Time
NUAA 15/3 Download 2010 Print 2010
CASIA-MFSD 50/3 Download(link failed) 2012 Print, Replay 2012
Replay-Attack 50/1 Download 2012 Print, 2 Replay 2012
MSU-MFSD 35/1 Download 2015 Print, 2 Replay 2015
MSU-USSA 1140/1 Download 2016 2 Print, 6 Replay 2016
Oulu-NPU 55/3 Download 2017 2 Print, 6 Replay 2017
Siw 165/4 Download 2018 2 Print, 4 Replay 2018
Experimental Comparison
Attack Presentation Classification Error Rate APCER
Bona Fide Presentation Classification Error Rate BPCER ACER =
(APCER+BPCER) / 2
Half Total Error Rate HTER
The HTER is half of the summation of the False Rejection Rate
(FRR) and the False Acceptance Rate (FAR)
Oulu protocols : OULU-NPU
a mobile face presentation attack database with real-world variations
[출처] : TDR/ FDR http://www.mohr-engineering.com/TDR_vs_FDR_Distance_to_Fault-A.php
Example
Thanks
Any Questions?
You can send mail to
Susang Kim(healess1@gmail.com)

[Paper] anti spoofing for face recognition

  • 1.
    Susang Kim(healess1@gmail.com) Anti-Spoofing Learning DeepModels for Face Anti-Spoofing: Binary or Auxiliary Supervision
  • 2.
    Anti-Spoofing 관련 아래논문을 바탕으로 설명 [논문] Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2018), Salt Lake City, UT, Jun. 2018 http://cvlab.cse.msu.edu/pdfs/Liu_Jourabloo_Liu_CVPR2018.pdf [Data] Spoof in the Wild (SiW) Face Anti-spoofing Database http://cvlab.cse.msu.edu/spoof-in-the-wild-siw-face-anti-spoofing-database.html
  • 3.
    얼굴인식 실제 서비스로이어지기 위해서는... 실재 얼굴과 Print,Phone(Galaxy,iPhone), 가면, 동영상등을 구분해야함 [출처] : http://biometrics.cse.msu.edu/projects/face_recog_spoofing.html
  • 4.
    By Auxiliary Supervision RGB이미지를바탕으로 Depth정보를 생성하고 rPPG정보로 분류
  • 5.
    What is rPPG? [출처]: https://www.noldus.com/facereader/remote-photoplethysmography-facereader
  • 6.
    Typical Application ofrPPG Desktop implementation of Remote Photoplethysmography – Measuring heart rate using facial video. (https://github.com/prouast/heartbeat)
  • 7.
    Proposed Method CNN으로 이미지정보를 학습 rPPG정보를 RNN으로 학습하여 구성
  • 8.
  • 9.
  • 10.
  • 11.
    Non-rigid Registration Feature Map과DepthMap 그리고 3D Shape의 3가지 Input => frontalized F output
  • 12.
    Spoof in theWild Database data from live and spoof viedos
  • 13.
    Anti-Spoof Data Datasets #of subj. / # of sess. Links Year Spoof attacks attacks Publish Time NUAA 15/3 Download 2010 Print 2010 CASIA-MFSD 50/3 Download(link failed) 2012 Print, Replay 2012 Replay-Attack 50/1 Download 2012 Print, 2 Replay 2012 MSU-MFSD 35/1 Download 2015 Print, 2 Replay 2015 MSU-USSA 1140/1 Download 2016 2 Print, 6 Replay 2016 Oulu-NPU 55/3 Download 2017 2 Print, 6 Replay 2017 Siw 165/4 Download 2018 2 Print, 4 Replay 2018
  • 14.
    Experimental Comparison Attack PresentationClassification Error Rate APCER Bona Fide Presentation Classification Error Rate BPCER ACER = (APCER+BPCER) / 2 Half Total Error Rate HTER The HTER is half of the summation of the False Rejection Rate (FRR) and the False Acceptance Rate (FAR) Oulu protocols : OULU-NPU a mobile face presentation attack database with real-world variations [출처] : TDR/ FDR http://www.mohr-engineering.com/TDR_vs_FDR_Distance_to_Fault-A.php
  • 15.
  • 16.
    Thanks Any Questions? You cansend mail to Susang Kim(healess1@gmail.com)