Abstract
Person recognition has been primarily studied with static images of faces. However, in real life we typically see the whole person in motion. This dynamic exposure provides rich information about a person's face and body shape as well as their body motion. What is the relative contribution of the face, body and motion to person recognition? In a series of studies, we examined the conditions under which the body and motion contribute to person recognition beyond the face. In these studies, participants were presented with short videos of people walking towards the camera and were asked to recognize them from a still image or a video that was taken on a different day (so recognition was not based on clothing or external facial features). Our findings show that person recognition relies primarily on the face, when facial information is clear and available. However, when facial information is unclear or at a distance the body contributes to person recognition beyond the face. Furthermore, although person recognition based on the body alone is very poor, the body can be used for person recognition when presented in whole person context and in motion. In particular, person recognition from uninformative faceless heads attached to headless bodies was better than recognition from the body alone. Additionally, person recognition from dynamic headless bodies was better than recognition from multiple static images taken from the video. Overall our results show that when facial information is clearly available, person recognition is primarily based on the face. When facial information is degraded, body, motion and the context of the whole person are used for person recognition. Thus, even though the face is the primary source of information for person identity, information from the body contributes to person recognition in particular in the context of the whole person in motion.
Meeting abstract presented at VSS 2017