Imagine that you are looking at a dog that is standing behind a picket fence. You do not see several slices of dog; you see a single dog that is partially hidden by a series of opaque vertical slats. The brain’s ability to join these pieces into a perceptual whole demonstrates a fascinating process known as amodal completion.

It is clear why such a tendency would have evolved. Animals must be able to spot a mate, predator or prey through dense foliage. The retinal image may contain only fragments, but the brain’s visual system links them, reconstructing the object so the animal can recognize what it sees. The process seems effortless to us, but it has turned out to be one of those things that is horrendously difficult to program computers to do. Nor is it clear how neurons in the brain’s visual pathways manage the trick.

In the early 20th century Gestalt psychologists were very interested in this problem. They devised a number of cunningly contrived illusions to investigate how the visual system establishes the continuity of an object and its contours when the object is partially obscured. A striking example of amodal completion is an illusion devised by Italian psychologist Gaetano Kanizsa. In one view, you see a set of “chicken feet” arranged geometrically. But if you merely add a set of opaque diagonal bars, a three-dimensional cube springs into focus seemingly by magic, the chicken feet becoming cube corners.

The astonishing thing is you do not even need to overlay real bars—even illusory ones will do. Here the otherwise inexplicable absence of boundaries terminating the chicken feet leads the brain to automatically infer the presence of opaque bars. So you see an illusory cube occluded by illusory bars!

The term “amodal completion” was coined to distinguish it from modal completion. Modal completion is the brain’s tendency to see the full outline of a nonexistent object, as occurs in Kanizsa’s classic triangle illusion. The brain regards it as highly improbable that some sneaky visual scientist has placed three black disks with pie-shaped wedges cut out of them precisely in this manner, preferring instead to see an opaque white triangle that is partially covering three black disks.

Note, however, that modal and amodal completion can coexist. For example, in the Kanizsa triangle, the brain amodally completes each disk behind the corners of the illusory triangle. Similarly, the illusory bars are modally completed, whereas the cube is amodally completed.

Peter U. Tse, a cognitive psychologist at Dartmouth College, has devised many elegant illusions to explore modal and amodal completion. One of them, shown in figure e, is ambiguous, as are many of our favorite illusions. There is a strong bias to see this figure as a stack of rings (amodally completed) encircling an opaque (modally completed) illusory cylinder. Yet one might have a very different take, seeing no cylinder and instead a column of C-shaped metal arches with the sharp ends facing forward. The bias toward seeing rings occurs because it better reflects the real world, which abounds in 3-D objects that occlude one another. Another of Tse’s illusions, which we fondly call “alien grabbing the last ­doughnut,” also has both modal and amodal aspects. It looks like a bunch of squiggles until the eye discerns a series of tentacular fingers coiled around a doughnut-shaped tube.

The Transparent Tunnel
You might think amodal completion involves reasoning (“there is a fence in the way, which is why I’m seeing slices of dog”), but in fact it is a perceptual phenomenon requiring no cogitation.

When you notice a wagging tail protruding from under the sofa and recognize that a dog must be attached, that is a logical inference. Whereas if the dog’s head were sticking out from the other side of the sofa, then in an automatic and effortless manner, via amodal completion, you would perceive a whole dog without actually seeing its hidden parts.

3 Comments

1. 1. jtdwyer 06:50 PM 8/30/10

   As I understand, the brain includes a specialized pattern recognition system that can recognize faces, familiar shapes, etc., much faster than we can process language, for example. It seem quite feasible that this 'dedicated hardware' is simply invoked many times in a brute force attempt to identify a discrete number of critical but incomplete shapes within visual data.
   
   General purpose computers serially process data, whereas parallel architectures can (once required data is in place) can perform a given function on an array of like data (such as an image) in the time required to process a single element (pixel). A specialized parallel image comparator could very quickly provide assessments of partial images, whereas a similar process implemented using serial functions would require an enormous amount of processing logic, move a lot of data and take a very long time to complete.
   
   This image recognition function is very important and useful to living organisms...

   Reply | Report Abuse | Link to this

2. 2. nthmost 08:48 AM 9/5/10

   I think it's likely we'll see the same generalized amodal completion algorithm at work in all forms of sensory processing and cognition: language, proprioception, etc.

   Reply | Report Abuse | Link to this

3. 3. nancedk 10:55 AM 9/17/10

   i am looking for special edition 169 best illusions i have called r customer service number it is disconnected please tell me where i can buy this issue or can i order it? dnance56@att.net

   Reply | Report Abuse | Link to this