Abstract
Person recognition depends on "telling people apart," and "telling people together" (Andrews et al., 2015). Observers must be able to determine that images of different people depict different identities and also that different images of the same person do not. Observers do this well with segmented and well-controlled images, but perform far worse with natural images. In particular, observers frequently label different images of the same person as different identities in an unconstrained card sorting task (Jenkins et al., 2011) - a failure to "tell people together." Presently, we investigated how both aspects of person recognition function when observers sort naturalistic faces and bodies in isolation, or presented together, for person recognition. We recruited 40 participants to sort multiple images of four unfamiliar individuals into groups based on identity. Participants were not told how many unique identities were present, and either saw images depicting faces only, bodies only, or a full photo. Across conditions, we examined how many groups participants made and their error rates for putting different people in the same group vs. separating images of the same person. Our participants substantially overestimated the number of identities in all conditions (Nface=18.0, Nbody=13.9, Nfull=15.2), and made significantly more face groups than body groups (p=0.006). While "Same-person/Different-group" error rates did not differ across our three stimulus conditions, "Different-person/same-group" error rates did differ, such that the full-image condition led to a significantly lower error rate than either face (p=0.029) or body sorting (p=0.004) and that body sorting led to a higher error rate than face sorting (p=0.024). This demonstrates that the ability to "tell people apart" differs as a function of face/body presence, but the ability to "tell people together" does not.
Meeting abstract presented at VSS 2016