Michael C. Mangini and Irving Biederman
University of Southern California

People reveal considerable expertise in the classification of a face in
terms of gender, expression, and identity, yet the representation
mediating such performance is often not available to conscious,
explicit description. To specify these representations, observers
classified faces appearing in sinusoidal noise as male/female,
happy/unhappy, or Tom Cruise/John Travolta. Unbeknownst to the
subjects, the underlying face stimulus was identical on every trial.
Therefore, all variations in the stimuli—and the subjects'
responses—could be attributed to the noise. The correlation of the
subjects' responses with the noise was used to compute a
"classification image" (Ahumada, 1996) that yielded clear exemplars
of the classes. Reverse correlation may thus provide a method for
making explicit otherwise ineffable perceptual representations.