Academic journal article Memory & Cognition

Unfamiliar Faces Are Not Faces: Evidence from a Matching Task

Academic journal article Memory & Cognition

Unfamiliar Faces Are Not Faces: Evidence from a Matching Task

Article excerpt

It is difficult to match two images of the same unfamiliar face, even under good conditions. Here, we show that there are large individual differences on unfamiliar face matching. Initially, we tried to predict these using tests of visual short-term memory, cognitive style, and perceptual speed. Moderate correlations were produced by various components of these tests. In three other experiments, we found very strong correlations between face matching and inverted face matching on the same test. Finally, we examined potential associations between familiar and unfamiliar face processing. Strong correlations were found between familiar and unfamiliar face processing, but only when the familiar faces were inverted. We conclude that unfamiliar faces are processed for identity in a qualitatively different way than are familiar faces.

People are remarkably poor at matching images of unfamiliar faces. For example, Figures 1 and 2 show arrays, reproduced from Bruce et al. (1999). Subjects were shown arrays of this type and told that the face of the person at the top might or might not be among the 10 targets. They were asked to decide whether or not that face was present and, if so, to pick the correct target.1 Performance on this task was surprisingly low. When the target was present, subjects picked the correct person on only about 70% of occasions. When the target was absent, subjects nevertheless chose someone on roughly 30% of occasions, despite knowing that half the arrays would not contain the target.

This level of performance is particularly striking, because the arrays used by Bruce et al. (1999) were constructed in some dimensions to optimize performance. All images were taken hi good lighting, from very similar fullface poses, and all were taken on the same day, eliminating transient differences in (for example) hairstyle, weight, or age. However, they were taken with different cameras, leading to different superficial image qualities. This has led some researchers (Hancock, Bruce, & Burton, 2000) to suggest that the task is carried out by subjects in a rather unsophisticated "image matching" fashion and that the change in camera can make this match difficult.

The difficulty of matching can also be observed in settings with different task demands. For example, when the Bruce et al. (1999) array task is reduced to a 10-alternative forced choice, so that all arrays contain the target, and subjects know this, performance still remains surprisingly low, at 79% (Bruce, Henderson, Newman, & Burton, 2001). In fact, the low levels of performance persist, even when subjects are shown two images and have to decide whether or not they are the same person (Henderson, Bruce, & Burton, 2001). Kemp, Towell, and Pike (1997) showed that people find it very difficult to match a passport-type photo to a real person, reporting very high error rates for supermarket checkout staff attempting to verify store cards bearing a photograph of the carrier. They concluded that security would not be enhanced by the introduction of such cards.

This poor performance for unfamiliar faces stands in contrast to our very good performance with familiar faces. For example, viewers familiar with targets can recognize them trivially in arrays such as shown in Figures 1 and 2 and can even do so with high accuracy in severely degraded images, such as those provided by poor-quality security surveillance cameras (Burton, Wilson, Cowan, & Bruce, 1999; Liu, seetzen, Burton, & Chaudhuri, 2003).

Here, we are primarily interested in unfamiliar-face processing and, particularly, individual differences in a facematching task. Although performance with unfamiliar faces is generally poor, there is nevertheless quite considerable variation among subjects. In the first experiment we report below, using the Bruce et al. (1999) matching task, performance ranged between 50% and 96%. It therefore seems reasonable to ask what factors might predict this performance. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.