Perceptual matching of identity between faces and video

Abstract

Recognizing people in naturalistic contexts requires an ability to extract identity-specific information from static and dynamic displays that may vary substantially in viewpoint and image resolution. The task is complicated further by the fact that videos contain information about both the identity of a person and the activity in which they are engaged. We tested participants' accuracy in matching people from video clips and static images presented side-by-side. On each trial, participants viewed a high resolution, frontal image of a face and a mid-range 10 s video of a moving person. The stimuli were pairs of males in their twenties, matched on peripheral features and build to make the task challenging. Participants were asked to determine if the identity of the people in the video and static images matched. The videos varied according to the activity of the person, with half showing the person approaching the camera and veering off to the left, and half showing the person at a distance engaged in a conversation. The probability of finding a match varied also, as high, medium or low, with match probabilities set to .50, .25, and .07, respectively. Participants were not informed of the target probability. Match accuracy (A') and response bias (C) were measured. Consistent with previous work, target probability affected response bias by loosening the criterion for the low target probability condition relative to the medium and high probability conditions. Surprisingly, target probability also affected match accuracy. We found a linear relationship between target probability and match accuracy, with the highest accuracy for the low target probability condition. No effects were found for the video type, walking versus conversation, on either match accuracy or response bias variable. These results are discussed in terms of their implications for eyewitness identifications that involve recognizing people across different image and video formats.

Footnotes

 Work supported by a grant from TSWG to A.O. and H.A.