Improving Fully Automated Landmark-based Face Morphing

A face morphing attack is an identity sharing scheme, where two individuals can use the same identity document. To do this, the attackers need to create a face image that resembles both of them using computer software. This image is called a morph and high quality morphs can fool both human officers and face recognition systems. High quality morphs are necessary for training and testing human officers, face recognition and morphing attack detection systems. However, manual morph creation is a tedious task which requires automation. In this paper, we present a fully automated morphing method which improves upon the state of the art by creating more realistic, ghost artifact-free morphs. Additionally, experiments with face recognition systems suggest that our morphs are equally or more likely to match with both attackers than current state of the art morphs with minor exceptions.

I. Introduction

Face morphing is the process of creating a composite face image (a morph) by digitally manipulating face images of different individuals, usually two. Under certain circumstances, the composite image looks like both contributors and can be used by one of them (accomplice) to apply for a new ID document. The other contributor (criminal) can then use the ID document to conduct illegal activities and evade justice. This identity sharing scheme is known as a Morphing Attack (MA) and is a serious and urgent security vulnerability, since it is difficult for both human officers and Face Recognition Systems (FRSs) to detect high quality morphs [1] [2]. An example of a MA is shown in Fig. 1. Fig. 1.

Example of a morphing attack with the accomplice on the left, the morphed passport image in the middle and the criminal on the right.