Temporal Convolutional Networks for Robust Face Liveness Detection

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Face authentication and biometrics are becoming a commodity in many situations of our society. As its application becomes widespread, vulnerability to attacks becomes a challenge that needs to be tackled. In this paper, we propose a non-intrusive on the fly liveness detection system, based on 1D convolutional neural networks, that given pulse signals estimated through skin color variation from face videos, classify each signal as genuine or as an attack. We assess how fundamentally different approaches – sequence and non-sequence modelling – perform in detecting presentation attacks through liveness detection. For this, we leverage on the Temporal Convolutional Network (TCN) architecture, and exploit distinct and TCN grounded types of convolution and architectural design schemes. Experiments show that our TCN model provides the best balance in terms of usability and attack detection performance, achieving up to 90% AUC. We further verify that while our 1D-CNN with a residual block variant performs on par with the TCN model in detecting fake pulses, it underperforms in detecting genuine ones, leading to the conclusion that the TCN model is the most adequate for a production environment. The dataset will be made publicly available to foster research on the topic. (https://github.com/novasearch/Mobile-1-D-Face-Liveness-Detection

Original languageEnglish
Title of host publicationPattern Recognition and Image Analysis - 10th Iberian Conference, IbPRIA 2022, Proceedings
EditorsArmando J. Pinho, Petia Georgieva, Luís F. Teixeira, Joan Andreu Sánchez
PublisherSpringer
Pages255-267
Number of pages13
ISBN (Electronic)978-3-031-04881-4
ISBN (Print)978-3-031-04880-7
DOIs
Publication statusPublished - 2022
Event10th Iberian Conference on Pattern Recognition and Image Analysis, IbPRIA 2022 - Aveiro, Portugal
Duration: 4 May 20226 May 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13256 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference10th Iberian Conference on Pattern Recognition and Image Analysis, IbPRIA 2022
Country/TerritoryPortugal
CityAveiro
Period4/05/226/05/22

Keywords

  • 1D convolutional neural networks
  • Liveness detection
  • Mobile presentation attacks
  • Temporal convolutional networks

Fingerprint

Dive into the research topics of 'Temporal Convolutional Networks for Robust Face Liveness Detection'. Together they form a unique fingerprint.

Cite this