Face classification

Human observers are more adept at classifying a face as, for example, male vs. female, or happy vs. unhappy, than they are at individuating that face (itself a form of classification). This result, as well as the finding of selective neurological impairment of face individuation in the condition known as prosopagnosia, gave rise to theories positing separate pathways for facial individuation and classification. However, all prior studies showing an individuation deficit suffered, minimally, from one of two confounds: (a) class uncertainty and (b) unequal stimulus differences. Here methodologies were developed that eliminate these confounds.;Observers perform a forced-choice match-to-sample task in which the similarity of the distractor is varied along a dimension (vector) defined by differences in gender, expression (happy-unhappy) or a pair of same gender, same expression individuals. The amount of physical stimulus energy required to achieve 75% threshold performance was measured. Normal subjects were most sensitive to expression differences and least sensitive to differences in individuals.;A prosopagnosic showed the same ordering and was within the normal range for the expression and gender discrimination tasks, but well outside of the normal range for the individuation task. What information do observers utilize to perform these tasks? This question was addressed by applying the classification image technique. This method uses an unbiased statistical process to define the information utilized in classifications. The classification images for each of three discriminations, gender, expression, identity, provided a linear estimate of the internal representations mediating face classification.;Finally, it is known that nonhuman animals can also make subtle discriminations between images. Do baboons trained on discriminating faces make use of the same information as humans? While their performance based on accuracy measures suggested they may be making use of similar facial information, the classification images showed they used information substantially different than the human observers.;Together the studies provide data about face classification that is quantitative rather than simply descriptive or ordinal. These findings show differing sensitivity to individuation and gender or expression classification; and different information selection between normal, prosopagnosic, and primate observers.