Experimental Evidence for Direct Perception: Persisting Layout (Gibson, 1979, Chapter 9)

We've now completed the first two sections of the book, which lay out the ecological approach. Gibson now turns to a review of the (at the time) existing empirical evidence for the various claims he has made. The first chapter focuses on how information supports the perception of a persisting layout of surfaces.

The first thing Gibson does is to contrast this with the study of depth perception. This is a venerable topic in vision (to this day!), because we experience a world that is spread out in this third dimension. For theories that begin with retinal images, science needs to find the ways that depth is recovered, and it focuses on the cues to depth present in images. Gibson notes that he tried using these theories during his work for the Air Force in WWII, and they failed to help guide pilot selection or training, at all. He refers to these theories as 'air theories', because they work to recover where objects are in space, and contrasts it to his approach, which he calls a 'ground theory', because it is about how we perceive surfaces arrayed along the ground. Information in the optic array for these surfaces replaces the cues to depth found in images. 

He also notes that the science of psychophysics is the wrong approach. Psychophysicists apply measured stimulation to the retina, and measure the behavioural response. The idea is that the forms used (simple shapes, Gabor patches, etc) are the elements of vision and so this is studying vision at it's most basic. Gibson rejects this idea - vision is not basic here, it is different. So the empirical evidence can't come from this kind of experiment; instead, you must make information available or not, and see how behaviour varies (hence the perturbation experiment). 

Surface vs No Surface

Gibson reviews several experiments that manipulate the presence or absence of the information for surfaces. He talks about Metzger's Gansfeld experiments, in which he placed people in rooms painted a uniform white. When the light was low enough to remove the information for the texture of the paint, people report seeing a fog or a haze. Gibson developed this technique with translucent balls in goggles, and described the perception as of '"nothing", in the sense of "no thing"' (pg 143). There is light, but no structure - no information, and so while light is detected, nothing is perceived. (A related effect is whiteout conditions, which are notoriously dangerous because perception fails this way.)

Perfectly clear glass is not perceived (as I can attest, having walked into two very clean doors in my life). We see the surfaces behind them by virtue of the information passing through unaffected. As the clearness is affected, so the surface comes into view and we perceive that. When there is structure in the light, we perceive what the light specifies. 

The optical tunnel is created by placing alternating black and white sheets of plastic in front of a person, with holes in the middle so you can see the sheets behind the front one. If you have enough sheets (creating a sufficiently dense array) and get the spacing right, people perceive a continuous tunnel receding into the distance (something you could roll a ball down). In some ways this is an illusion; but for Gibson, it was about manipulating the structure of the optic array, with predictable effects on perception. 

Surfaces of Support

The next experiments vary the perception of surfaces as affording support. He describes studies with glass floors (e.g. the visual cliff) and talks about how when babies and animals encounter these, they behave in ways that are less about depth perception and more about perceiving affordances. He also describes another illusion, where an object is raised about a textured surface and is seen as being at the distance specified by where it cuts the texture; this speaks to the importance of the ground. 

The Ground As Background

Instead of studying depth perception from image cues with images, Gibson talks about studies of the perception of surface layout using actually laid out surfaces. He talks about experiments in which size is correctly judged at varying distances, even though the various pictorial cues (e.g. retinal image size) are changing. Participants could bisect distances both from themselves to a target, or between two targets, when there was a textured ground, and he discusses the informational basis for this. He notes the importance of the horizon, and the fact it is optically specified in many situations, and that this fact of optics helps calibrate many features of vision. 

The Perception of Slant

Gibson had previously studied how a texture gradient could serve as a cue to slant; by painting a gradient on a vertical surface. People reliably underestimated the slant, as measured with a palm board. The problem, Gibson now realises, is that the optical slant was not properly specified as relative to the ground, and also that people were also perceiving the vertical surface. 

Evidence Against the Direct Perception of Surface Layout

Gibson discusses two examples from Ames, the distorted Window and the distorted Room. These are examples of the argument from equivalent configurations; the idea that you can present the retina with forms that are the same, even though the underlying source is the same. Given this 'fact' of vision, additional processes seem to be demanded. Of course, the illusions dissolve when you are allowed to move from monocular, peephole vision and explore with two eyes over space and time. Runeson (1988) lays this all out with exquisite detail (blogged here). 

Conclusion

Gibson concludes that there are empirical demonstrations of the fact that we perceive a layout of surfaces as specific by information, if and only if that information is there. This is all preliminary evidence in favour of replacing image based depth perception with information based surface layout perception.