What can Hollow Faces tell us about Visual Perception?

The Hollow Face Illusion

A concave mask or mould of a face appears convex when viewed from beyond a certain distance, and appears to follow us when we move. This is the hollow face illusion (Figure 1).

Figure 1: Four views of a hollow mask

The hollow face illustrates how we can use illusions to address important questions in perceptual psychology, for example:

1. How do we perceive three-dimensional space and shape?

2. What prior assumptions do our brains employ in interpreting the visual world?

3. Do our brains process faces in a special way?

The Ambiguity of Shape-from-shading

All the normal sources of depth information are present in the hollow face, including shading, motion, stereoscopic disparity and texture.

Shading itself is fundamentally ambiguous: lighting a convex surface from one direction or a concave surface from the opposite direction produces the same pattern of shading.

Figure 2: Shape from shading

When you turn Figure 2 upside-down your perception flips, so that what appeared convex appears concave, and vice versa. It appears that the perception of concavity versus convexity is governed by two implicit assumptions: (1) lighting is coming from a single source; (2) the source is located above rather than below.

Perception as Hypothesis

In the hollow face illusion, depth information from other sources such as stereoscopic disparity and texture could disambiguate concavity and convexity. However, this information is over-ridden by the brain's prior assumptions about the world, demonstrating that it is not just the perception of fundamentally ambiguous stimuli that is influenced by prior assumptions.

Thus, we can think of perception as hypothesis testing: a process of matching bottom-up sensory information with top-down expectations about the world. However, what exactly constitutes the "top" is unclear: the hollow face illusion persists even when we know that we are looking at a hollow mask.

When we move relative to the mask, the changes in the retinal image are also ambiguous. One side of the face progressively disappears from view as we move around it while the other side gets revealed. The changes in the image of a static concave face are the same as those that would be produced by a convex face turning to follow our movement. For our perception of a convex face to remain consistent with dynamic occlusion cues requires that the face be seen as turning so as to be self-occluding.

Do our visual systems process faces in a special way?

The hollow face illusion is robust to changes in lighting direction, suggesting that the expectation for faces to be convex outweighs the expectation for lighting to come from above. The existence of an analogous hollow potato illusion shows that the visual system expects objects in general to be convex, not just faces, but the illusion is stronger for faces.

The hollow face illusion disappears when we are too close, typically at viewing distances of one to two metres, as the depth cues in the retinal image become harder for our visual systems to ignore. The strength of the illusion can be compared for different objects by measuring the distance at which the illusion disappears as we approach different hollow objects. Short distances imply a strong illusion.

Figure 3: A comparison between the hollow-face and "hollow-potato" illusions (Hill & Bruce, 1994)

Hill & Bruce (1994) found that upright hollow faces produce stronger illusions than hollow potatoes or inverted hollow faces, suggesting that object-specific knowledge is important in generating the illusion. The inversion effect is stronger for faces than for other familiar objects with a cardinal orientation, suggesting either: (1) faces are processed in a special way; (2) faces lie at the extreme of a continuum of familiar objects. To distinguish between these two possibilities we must look to sources of information beyond hollow objects.

Reading:

Gregory RL (1998) Eye and brain. Oxford University Press, Oxford, UK, pp. 207-210.

Hill H & Bruce V (1994) "A comparison between the hollow-face and 'hollow-potato' illusions", Perception 23, 1335-1337.

Ramachandran VS (1990) "Perceiving shape from shading", in I. Rock (ed.) The Perceptual World, WH Freeman, New York, pp.127-138.