Skip to main content
Open Access Publications from the University of California

How do expertise and realism moderate the boundary between real and digital faces?

  • Author(s): Day, Jennifer Anne
  • Advisor(s): Davidenko, Nicolas
  • et al.
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Public License

I explore a new approach to studying the development of face expertise by examining digital representations of faces. Typically digital faces have lower recognition scores than real faces, and I propose expertise and style as factors that moderate this boundary. Experiment 1 recruited participants who have played over 50 hours of The Elder Scrolls V: Skyrim as an expert population (experts, n=51) and compared their score on an upright and inverted face recognition task against participants who had not played Skyrim (novices, n=55). We also tested two different races of faces from Skyrim (Nord and Altmer). Participants performed significantly better on upright faces, t(104) =14.056, p <.0001 and Altmer faces, t(104) =-5.346, p < .001, and experts performed significantly better than novices, t(103) =2.664, p < 0.01. Participants then watched a video of gameplay with eye-tracking where participants’ proportions of fixations on Skyrim faces were recorded. This face proportion measure did not significantly correlate with any other measure. Finally, participants completed a survey with questions about the number of hours played, video game habits, open-ended questions about experiences in Skyrim, and a character recognition task. Experiment 2 recruited a new population of novice participants (n=46) and compared scores on an upright and inverted face recognition task for morphed photos of human faces to 3 different styles of video game faces that ranged from highly realistic (Monster Hunter: World), moderately realistic (Skyrim), to highly stylized (Blade & Soul). Participants performed significantly better on upright faces over inverted faces, F(1, 45) =13.06, p <.0001 and on realistic faces over stylized faces, F(1, 45) = 54.657, p < .0001. Participants demonstrated a significantly larger upright advantage for stylized faces over realistic faces F(1, 45) = 11.97, p < .01. Participants then completed a survey with questions about video game habits, perceived task difficulty, and looking strategies. The results for experiments 1 and 2 provide evidence to suggest that both expertise and style can account for reduced performance in digital faces, and open up questions about how these factors interact.

Main Content
Current View