Algorithmic Fairness in Face Morphing Attack Detection

Face morphing attacks can compromise Face Recognition System (FRS) by exploiting their vulnerability. Face Morphing Attack Detection (MAD) techniques have been developed in recent past to deter such attacks and mitigate risks from morphing attacks. MAD algorithms, as any other algorithms should treat the images of subjects from different ethnic origins in an equal manner and provide non-discriminatory results. While the promising MAD algorithms are tested for robustness, there is no study comprehensively bench-marking their behaviour against various ethnicities. In this paper, we study and present a comprehensive analysis of algorithmic fairness of the existing Single image-based Morph Attack Detection (S-MAD) algorithms. We attempt to better understand the influence of ethnic bias on MAD algorithms and to this extent, we study the performance of MAD algorithms on a newly created dataset consisting of four different ethnic groups. With Extensive experiments using six different S-MAD techniques, we first present benchmark of detection performance and then measure the quantitative value of the algorithmic fairness for each of them using Fairness Discrepancy Rate (FDR). The results indicate the lack of fairness on all six different SMAD methods when trained and tested on different ethnic groups suggesting the need for reliable MAD approaches to mitigate the algorithmic bias.

1. Introduction

Face recognition systems are widely deployed in various security-based applications, including border control. While FRS can provide a user-friendly, seamless, and reliable verification solution, they are also known to be vulnerable to different types of attacks, including the recently reported face morphing attacks [41, 9, 16]. Face morphing attacks against FRS are carried out by combining face images of multiple subjects to generate a single image that can be verified equally against all contributing subjects [41].. Existing works on face morphing attacks have reported the vulnerability of FRS towards such attacks [23, 30, 12, 22, 21, 19, 32, 24, 25, 17, 31, 38, 39, 33, 20, 42, 26, 43, 34, 35, 7, 7, 2, 3]. Extensive benchmark presented by NIST FRVT MORPH [20] and the Bologna Online Evaluation Platform (BOEP) [26] also indicate the high vulnerability of FRS with respect to face morphing attacks. In addition to FRS, also human observer analysis (including forensic experts, border guards and biometric researchers) has also reported the challenges to detect face morphing attacks [14, 28, 15, 27]. These factors have motivated the development of multiple automatic face Morphing Attack Detection (MAD) to detect the evolving morphing attacks, especially tailored for border control applications. A number of MAD techniques have been proposed in the past five years [7] [2] [25] [34] [41] [38] [22] which can be broadly classified as Single image-based MAD (S-MAD) and Differential based MAD (D-MAD) [41]. The S-MAD techniques use a single image on which morphing detection is performed and is deemed suitable for forensic applications.