Abstract
In television, movies, and video games, we are exposed to a diversity of human faces ranging in naturalness. Some faces appear entirely human, other faces appear entirely artificially generated, while still others fall along a continuum between these two extremes. Encoding the average naturalness of crowds may facilitate a more/less immersive media experience. In order to determine whether viewers are able to encode the naturalness of faces in a brief glance—we choose 100 face stimuli varying in naturalness. Participants from Amazon Mechanical Turk viewed each face for 1 second. Once the face disappeared, participants rated each face on a Likert scale, with 1 representing the most natural-appearing face and 10 representing the least natural-appearing face. The raters were highly reliable in their judgements of naturalness, and we used their ratings as a baseline for the next experiment. In Experiment 2, we generated groups of faces from the stimulus set. Each group was comprised of 6 faces and the average naturalness of the crowds varied from 3 to 7 on the Likert naturalness scale. New participants from Amazon Mechanical Turk participated in this experiment. On each trial, participants viewed the crowds of faces for a limited exposure duration. Once the crowd disappeared participants responded using the Likert scale to report the average naturalness of the crowd. We used a bi-variate correlation to compare participants' judgments of the average naturalness of the crowd of faces to the baseline ratings taken from independent observers in Experiment 1. The judgements from the two independent groups were significantly correlated, strongly indicating that participants are able to perceive the ensemble naturalness of crowds.
Meeting abstract presented at VSS 2018