On the perceptual structure of face space

We are able to recognize very many different faces of individuals we know, apparently using a complex and ill-understood set of identifying features; it seems natural to assume that faces are perceived as spanning the equivalent of a high-dimensional vector space. I explore ways to probe the structure of perceptual face space without making a priori hypotheses about either the space itself or the mechanisms of perception and recognition, and using solely neuronal responses recorded in the monkey, and metrics derived from their mutual similarities. Within this approach, the dimensionality of face space remains an elusive concept, but the metric content and ultrametric content of the face sets used can be quantified and compared with those of other perceptual sets.