Saturday, 16 September 2017

The Information for Progressive Occlusion

Gibson's ecological psychology is weird, if you are coming from a more traditional information-processing background. The two approaches make radically different assumptions about the nature of the world to be perceived; they have radically different ontologies. This means that there is little if any useful overlap in the way they do things, and communicating across the gap is very hard. I have a recent paper - preprint here - where I go into detail about the two ontologies as I defend ecological psychology from interface theory. It's essentially Turvey et al, 1981, but that's a bear of a read if you aren't already ecologically minded. Do mine first :)

Anyway, concrete examples help. My go-to is the outfielder problem but people are tired of that one. My other favourite is progressive occlusion (Gibson, Kaplan, Reynolds & Wheeler, 1969; Kaplan, 1969). Gibson worked this example up himself in great detail and so it stands as a nice concrete example to illustrate some elements of the ecological ontology. Given the recent total solar eclipse, it seems like the right time to blog it!

This post will review occlusion, talk about how it works and work with some demos. These are all linked from here; there is Matlab/Psychtoolbox code to run a demo, a video of that running and a Powerpoint with some slides. I'll refer to these throughout - occlusion is a dynamic process and so you need to see it moving for it to make sense.

Progressive Occlusion
Occlusion is when one thing is behind another, and the front object blocks your view ('occludes') of the rear object. Progressive occlusion is when you see one object moving in front of another; this is the case at hand. 

In the Powerpoint demo, show the slide with the animation. What you will see is a single textured square; then a smaller textured square will begin to move in a circle. What you should experience is the two surfaces separating in depth (i.e. there should seem to be a gap between them) with the small square in front of the large one. So long as it is moving, you will see the small square but shortly after it stops it will fade from sight. 

Gibson & Kaplan's Information Analysis
The information question is, what is happening to optical texture in the optic array over time? How, exactly, is the optic array projecting to the eye changing? 

As the small square moves, it creates a small patch of local optic flow that is moving differently from the rest of the array. This square patch of optical texture progressively occludes the optical texture from a region of the larger square. That texture goes out of sight; it vanishes from the optic array. But it does so in a very particular way.

The motion of the small square defines one edge as leading and one as trailing. At the leading edge, optical texture from the large square is progressively deleted, that is, it disappears over time. At the tailing edge, optical texture from the large square is progressively accreted, that is, it accumulates over time. Along the two side edges the local flow moves differently from the larger array but there is neither accretion nor deletion of optical texture. (The exact details will vary depending on the front surface shape.)

It is this higher order relational pattern in the optic array that specifies progressive occlusion and distinguishes it from disappearing or going out of existence.

Some Ontological Things
One of Gibson's key weird arguments is that sensations and retinal images are not the basis of vision. What matters is information, those higher-order relational patterns. Occlusion shows this nicely. 

  • On the retina: optical texture is disappearing and going out of existence at the leading edge and appearing and coming into existence at the trailing edge. The light rays from the occluded part of the surface are literally gone from the retina and then return with no clear way to know they are rays from the same part of the surface. 
  • The higher order informational pattern is as I described above; a regular pattern of deletion and accretion of optical texture, the timing of which is linked by the size, shape and speed of the occluding object.
  • In perceptual experience: one surface passes in front of another surface, and the latter continues to exist in its unoccluded form. The surface is not disappearing, it is being occluded.

Perception does not match sensation, but it does match information. 

Second, the informational pattern is specific to progressive occlusion, and it is a lawful consequence of the dynamical nature of the occlusion event. Only progressive occlusion produces this particular relational pattern, and that particular relational pattern is a lawful, necessary consequence of what is happening in the world. If you have one surface occluding another and an optic array, this relational structure is in there for you to use. If that structure is there in the optic array for you to use, then out in the world, one surface is progressively occluding your view of another. This is the Turvey et al (1981) symmetry principle that underpins direct perception; see this post and my interface theory paper for more detail. 

So here we have many of the key features of the ecological approach to perception. Perception is based on higher order relational information, not sensations. Information is lawfully related to the underlying event dynamics and thus can specify (via symmetry) that event in a way sensations cannot. And when you test observers, their perceptual experience is best explained by using lawful, invariant information and not the wildly variable and non-specific sensations. 

Some Empirical Evidence
Another reason I love progressive occlusion is that I get to use work by Zenon Pylyshyn and one of his students, Brian Scholl, in support of the Gibsonian analysis. Petty, perhaps, but these things can be fun :)

Scholl & Pylyshyn (1999) used a multiple object tracking through occlusion paradigm to investigate what it means for an object to be considered a persisting object, even when it goes briefly out of view. 

They tested participants with three displays; link to videos on Scholl's webpage

  1. Baseline: 4 objects flash, then everything starts moving, all remain in view throughout. Your job is to fixate the red cross and track the 4 objects that flashed. This is pretty easy.
  2. Occlusion: same set up, except now the 4 objects can pass under a couple of black bars in the display that you can't see. They go out of view and the visual information specifies progressive occlusion. 
  3. Implosion: same as Occlusion except when the objects go out of view the boxes get smaller and smaller, then come into view by getting bigger and bigger
The basic result (Figure 1) is that objects remain objects in the baseline and occlusion conditions but not the implosion condition. When people view the implosion displays, the visual system effectively 'lets them go' as no longer existing, so when they reappear they are considered to be new objects and tracking fails. During occlusion, the visual system 'keeps ahold' of them and considers them to be still there, just out of view, and when they come back they are treated as the same object. 
Figure 1. Data from Scholl & Pylyshyn (1999). The implosion condition performance is essentially at chance.
Progressive occlusion is when one object passes in front of another and blocks your view of that object over time. Optically this causes texture elements in the array to be deleted at the leading edge and to accrete at the trailing edge. This particular relational pattern in the optic array is lawfully related and specific to a progressive occlusion event. Despite optical elements going in and out of existence, what is perceived is the continued existence of the occluded object; perception is based in the time-and-space-extended information and not on the retinal image. When an object is perceived to have been occluded, it is perceived to continue to exist and so object based attention will, at least for a while, continue to treat it as an object and can find it again when it comes back into view. When the object is perceived as having gone out of existence, this is no longer possible. 



  1. Did Bingham ever show you the video he made while carrying a video camera down a hallway? It's another good example of information about something (walking person) in the absence of any image of that something.

    1. Ha, yes :) I TA'd his undergrad course and sat his graduate course, so I got it at least twice :)