Multi-Perspective Dynamic Features for Cross-Database Face Presentation Attack Detection

Fanghao Liu,Yukun Ma,Yaowen Xu

doi:10.1109/access.2020.2971224

Abstract

With their growing popularity and widespread applications, face recognition systems are attracting more attention from attackers. Thus, face presentation attack detection has emerged as an important research topic in recent years. Existing methods for face presentation attack detection are affected by different cameras and display devices, and their performance is degraded in cross-database testing. In this paper, we propose a face presentation attack detection scheme that fuses multi-perspective dynamic features. One feature is the globally extracted temporal motion pattern of a face in a video. This involves mapping the local and global motion information of the face in the video into a single image. The motion patterns of genuine and fake faces are different, and these patterns are independent of cameras and display devices. Another feature is the visual rhythm of noise patterns, which differs significantly between single and secondary imaging. The proposed scheme fuses these two features at the decision level. Cross-database tests were conducted among the CASIA-FASD, MSU-MFSD and Replay-Attack databases. The experimental results show that the proposed scheme outperforms state-of-the-art algorithms.

Highlights

In recent years, the development of deep learning technology has significantly improved the accuracy of face recognition and facilitated widespread applications of face recognition systems
Yang proposed a face presentation attack detection algorithm based on a convolutional neural network (CNN) and obtained a half total error rate (HTER) of 2.93% for the test set from the database that was used as the training database; the HTER increased to 39.49% [11] in the cross-database test
To improve the cross-database performance of face presentation attack detection, we propose a novel method for extracting temporal-global facial motion patterns

Summary

INTRODUCTION

The development of deep learning technology has significantly improved the accuracy of face recognition and facilitated widespread applications of face recognition systems. It is obvious that these features present the local rather than global motion patterns of a video These algorithms are mainly tested on a single database, in which the facial images, both real and fake, are generally captured using the same device. Yang proposed a face presentation attack detection algorithm based on a convolutional neural network (CNN) and obtained a half total error rate (HTER) of 2.93% for the test set from the database that was used as the training database; the HTER increased to 39.49% [11] in the cross-database test This is because the final imaging results of genuine and fake faces are influenced by the cameras, display devices, and imaging environments. On the basis of these observations, Pinto proposed a noise signature to distinguish real from fake biometric data and achieved good performance [19] Motivated by these factors, we propose multi-perspective dynamic features for face presentation attack detection.

RELATED WORK

CLASSIFICATION

Findings

CONCLUSION