Information useful for identifying a person can be found both in the face and body. Previous studies indicate that when an entire person is visible, we rely strongly on the face for identification, even if the body can be useful. In this study, we measured the utility of the face versus body for identification, using images of people that varied in the quality of identity information in the face. Face quality was varied using similarity scores generated by state-of-the- art face recognition algorithms from an international competition. These algorithms estimated the similarity of faces in a large set (> 1,000,000) of image pairs that showed people, including the faces and top half of the body. Using these similarity scores, image pairs were stratified into three groups, representing good, moderate, and poor performance for the face recognition algorithm. In this study, participants matched identity in image pairs sampled from the three groups, using versions of the stimuli edited digitally to show only the face or body. Consistent with the algorithm stratifications, performance with the face declined across the three conditions. The face supported more accurate identification than the body in the good and moderate conditions. In the poor condition, performance from the face and body was comparable. Using data from a previous study, we compared the face- and body-only identity judgments to judgments based on the original image. The original unedited image supported the best overall performance in the good and moderate conditions. Notably, performance in the poor condition was equivalent for the face, body, and original images. The results indicate that in poor viewing conditions, the body may support identification decisions that are as accurate as those made from the face or from the entire person.
Citation: Applied Cognitive Psychology
Pub Type: Journals