Finding the Suitable Doppelgänger for a Face Morphing Attack

Abstract

ID cards are uniquely linked to one individual via a printed or electronically provided facial image. Even though the face is treated as universal and distinctive characteristic, twins can weaken this distinctiveness because of their biological similarity. Also, humans might falsely recognise an unknown person as a friend - colloquially named a Dop-pelgänger. Recently it was demonstrated that this biological effect of similar data subjects can be purposefully established between two individuals in order to improve the vulnerability of the so-called morphing attack. This image manipulation technique creates a melted facial image which is similar to two or more data subjects. If embedded into an ID card, the manipulated reference image can be used by all participating individuals and thus the concept of a unique link is broken. This work elaborates the rather neglected part of selecting morph pairs based on a similarity score instead of a simple random assignment. It discusses the applicability of different possible algorithms. The finally developed approach considers complex real-world constraints while being executable in a reasonable amount of time and producing acceptable large morph sets. It is shown that this algorithm greatly increases the vulnerability of automated face recognition systems. Surprisingly, it also proves that an effective pre-selection of pairs questions the need of in-depth optimized morphing algorithms.