Alexander Röttcher defended his Master Thesis on: Finding the suitable Doppelgänger for a Face Morphing Attack

Passports, membership cards, or other ID cards often have one thing in common: A
facial image is printed on the card and electronically provided on a chip to allow a
verification if the card belongs to a person. The face is a universal characteristic which
everyone has, can be easily captured by any camera and has a high acceptance to be
used for the identification purpose. However, twins destroy or at least weaken
the distinctiveness of different faces due to their biological similarity. Also, one might
have experienced to falsely identify an unknown person as a friend – colloquially
named someone’s Doppelgänger. Can this biological effect of similar data subjects be
purposefully established between two individuals in order to share an identity?
In this work, the so-called morphing attack, firstly described by Ferrara et al.,
is further analysed. This image manipulation technique creates an artificial facial
image which is similar to two or more different individuals. If introduced into
an ID card, this manipulated reference image can be linked to all participating
individuals. While the attack itself is rather simple to perform, a recent survey
concludes that state-of-the-art research results are difficult to be compared to each
other. This work presents a first step-by-step research guideline, provides ways for
automation, and discusses the impact of different decisions in order to tackle this
challenge. By considering not only the morphing process itself but all preparatory
work and subsequent processes, it permits a global view on how odds-and-ends can
greatly impact reported performance results. Especially, it is shown that a currently
rather neglected part, the pairing of similar data subjects from which facial images
are morphed together, negatively influences the vulnerability of automated face
recognition systems as well as the morphing attack detection performance. Therefore,
a new pairing algorithm is developed which considers complex real-world constraints
while being executable in a reasonable amount of time.