© Fraunhofer IGD
© Fraunhofer IGD
© Fraunhofer IGD

No, we are not talking about amusing apps that swap our relatives’ faces around, superimpose cute puppy-dog masks or display our unmistakable resemblance to a Hollywood star. When performed correctly, face morphing can pose a serious threat.

A single passport, multiple identities

Face morphing can be employed to assign multiple identities to a single passport photo. To morph photos of faces, the people depicted do not have to be related: a certain degree of similarity is sufficient. The artificial intelligence behind biometric facial recognition systems will recognize the successfully shared and merged passport photo as belonging to multiple people and will let them all through, with one and the same passport. This is possible if the facial nodal points that are key to recognition are perfectly aligned. And human intervention is not a solution either—the naked eye is unable to detect a morphed passport photo.

Learning from fraudsters

Cyber attacks are a major problem; ideally, they can be thwarted at an early stage, before their impact is felt. However, the “bad guys” are often one step ahead, identifying vulnerabilities before developers are aware of them. For researchers at Fraunhofer IGD, that means turning their hands to the fine art of morphing. Using the MorGAN (Morphing through Generative Adversarial Networks) image generation methodology, we have created a database with more than 1000 morphed photos. Traditional solutions for the identification of morphing attacks have not been able to successfully pinpoint MorGAN images. The next step is for the scientists to find a way of outwitting their own devious technology.