Scientists publish datasets and tools for detecting face morphed identity documents
A research paper published this month has tools to neutralize face morphing attacks, which use an altered biometric reference image in an ID document.
Four researchers from the Swiss Idiap Research Institute have provided, via IEEE, two datasets and tools for four kinds of morphing attacks against biometric systems. The scientists say efforts to detect these morphing attacks have been slowed because of a lack of relevant datasets and tools.
Germany two years ago banned face morphing in an attempt to prevent multiple identities from being attached to one altered image.
Two so-called classical types rely on facial landmarks based on OpenCV and FaceMorpher. The other two use StyleGAN 2 to create synthetic morphs from generative adversarial networks.
The team also analyzed the vulnerability of four facial recognition algorithms that they consider state of the art: FaceNet, ISV, ArcFace and VGG-Face.
At issue is the reliability of security systems — such as border and access control — that increasingly use face biometrics. In 2014, an Italian team of researchers showed that it was possible for, say, a wanted criminal to use a morphed photo to travel as someone who is not being pursued.
The researchers expressed surprise that despite creating images with “higher visual appeal,” GAN-based morph attacks are less of a security threat than are classical morphs.