Group Decision Performance of Human Examiners for Identity-aware Face Image Manipulation Detection

Face image manipulation technologies have become popular in recent years, allowing people interested in committing fraud to access services undetected. Although the research community keeps working on improving automated attack detection algorithms to detect face image manipulations (e.g. morphing and face swapping), human experts working in migration and law enforcement remain ultimately responsible for verifying a person’s identity and the authenticity of their documents. Therefore, it is essential to assess human capabilities in detecting manipulated face images, especially in identity-aware differential scenarios where an examiner must compare a trusted live face image taken in an uncontrolled environment with a portrait stored in a database or printed in an official document.