Avatars versus the people: Photo-realism, spontaneous detection and trait inferences of digitised faces.

Abstract

Technologies aiming to imitate human faces are becoming increasingly realistic. This study investigates a facial imitation technology that is becoming widespread - digital characters of people for presentation in virtual reality. Avatar faces were created from high-resolution 3D scans of real people. Across a series of four experiments, the photo-realism of these avatar faces was compared with passport-style face photographs of the same persons. In Experiments 1 and 2, these stimuli could be distinguished with high accuracy when a direct comparison of avatars and photographs was possible. In contrast, discrimination accuracy decreased when avatars and photographs were encountered in isolation, while awareness that avatar faces had been encountered was also low. Experiments 3 and 4 showed that avatars and face photographs generate similar trait inferences of attractiveness, dominance and trustworthiness. In cases where differences between avatars and photographs emerge, analysis of viewing patterns indicates that these originate from the eye region of these stimuli, which receive more fixations in avatars than face photographs. These findings demonstrate that the visual realism of avatars can closely resemble that of face photographs, particularly in contexts in which realism is not explicitly evaluated. Differences between avatars and photographs become more apparent when participants are cognizant and able to make direct comparisons.