Human ability to match synthesised faces to their constituent faces

Abstract

Currently, facial recognition systems are used widely within various government agencies as a means of identity verification, and these systems involve a human operator in the final decision making process. Previous studies have shown that both face recognition systems and humans are vulnerable to variables that may impede the face matching process, including the use of morphed images, which are created by digitally combining multiple constituent faces into a new face. Therefore, the current study aims to further investigate how the usage of different types of morphed faces can affect human face matching performance. Participants (N = 51) from the University of Adelaide and the general public completed 112 computer-based one-to-many face matching trials in a repeated measures design. The type of morphed face used as the target image varied for each trial, and was either made from 2, 8 or 16 constituent faces, or was a control non-morphed face. Results indicated that the usage of 8-Image morphs resulted in significantly higher accuracy and confidence, as well as faster response latency. Future research could be conducted using morphs made from similar faces, and employ multidimensional scaling methods to map the morphs and their constituent faces in face space.