Evidence for holistic processing of facial age

Abstract

A composite face, produced by pairing the top and bottom halves from two different faces, produces a strong percept of a “new” face: It is difficult to identify either half in isolation. Three experiments examined whether a similar “composite face effect” (CFE) occurs with age judgements. Composites contained halves from faces of different ages. Participants tried to judge the age of the top half, ignoring the bottom. Experiment 1 found that age estimates for the composites’ top halves were biased towards the bottom halves’ chronological ages. Inverting the stimuli attenuated this effect but did not eliminate it entirely. The CFE for age perception is not based on skin texture information, since it persists when texture is eliminated by blurring or random noise (Experiments 2 and 3). Methodological differences preclude too detailed a comparison with other CFEs, but possibly the CFEs for age, recognition, and expression perception are all based on a common initial “holistic” encoding phase in face processing.

Keywords:

• Age perception
• Chimeric faces
• Composite face effect
• Configural processing
• Face perception
• Holistic processing

Notes

¹Use of the terms “holistic” and “configural” is rather inconsistent in the literature on face processing. Tanaka and Farah (1993) suggested that faces evoked “holistic” processing, in the sense that the visual system processes faces as indissoluble wholes, rather than as collections of isolated features. The term “configural” processing has been used to refer to a form of processing that makes use of the fine-grain spatial interrelationships between facial features, their so-called “second-order relational” properties. The experiments reported here are concerned with whether age processing involves the obligatory extraction of information from the entire face, or whether participants can attend selectively to one part of the face or another; thus, the term “holistic” processing is probably most appropriate in the present context.

²As with the configural/featural distinction in face recognition, this trichotomy between “featural”, “configural”, and “surface” cues breaks down if pushed too far. For example, changes to the nose and lips are likely to produce changes to the overall configuration of the face, and overall textural changes may affect the appearance of some features more than others—sagging due to a loss of tissue elasticity is more likely to affect the eyes than the nose because the latter has better structural support.

³Using a Minolta LS-100 lightmeter, mean luminance values were computed for each face, based on an average of four readings (centre of forehead, chin, left, and right cheeks). For the three “young” faces, the means were 38, 49, and 61 cd/m². For the three “medium” faces, they were 22, 42, and 75 cd/m². For the three “old” faces, they were 18, 46, and 61 cd/m². (All readings were taken under normal viewing conditions, i.e., with the room's fluorescent lighting on.) Note that although luminance values varied from face to face, they provided no systematic cue to age.

⁴As one of our reviewers pointed out, it would have been preferable to separate the face halves with a thin horizontal line, to clarify where one half ended and the other began. We agree, but we do not believe this posed serious difficulties for participants in practice. As can be seen from Figure 1, a number of factors helped to make it fairly easy to distinguish between the upper and lower face halves. First, there were minor differences in image brightness; second, there were individual variations in facial complexion; and third, we made no attempt to hide the fact that the composites were comprised of different faces (alignment between the two halves was not perfect in all respects—look at the noses in Figure 1E, 1F, and 1G) None of the participants expressed any confusion over where one half ended and the other began, either during the experiment or afterwards when being debriefed.

⁵The mean luminance values for the faces remained very similar to those in Experiment 1. For the three “young” faces, the means were 36, 50, and 59 cd/m². For the three “medium” faces, they were 20, 41, and 73 cd/m². For the three “old” faces, they were 16, 44, and 60 cd/m².

⁶As one of our reviewers pointed out, the fact that this level of blurring impairs face recognition does not necessarily mean that it is sufficient to remove textural cues to age. Visual inspection of our stimuli suggested that skin texture information appeared to have been satisfactorily removed by our manipulations, but it would be desirable to conduct a more systematic study of the effects of various levels of blurring on age estimation.

⁷For the three “young” faces, the mean luminance values were 67, 79, and 86 cd/m². For the three “medium” faces, they were 64, 89, and 108 cd/m². For the three “old” faces, they were 63, 91, and 98 cd/m². The “texture” procedure markedly increased the faces’ mean luminance values, compared to Experiments 1 and 2.