New Book Explains Age-Old Mystery Of Geometrical Illusions

In a new book " Perceiving Geometry: Geometric Illusions Explainedby Natural Scene Statistics" (Springer), Purves and colleague CatherineHowe explore why the brain generates geometric illusions.

Visual perception is a daunting task for the brain, explains Purves,because light streaming into the eye carries only ambiguous informationabout the environment.

"The basic problem, recognized for several centuries, is that theimage on our retinas can’t specify what’s out there in the world," saidPurves. "The light received by our retinal receptors tangles upillumination, reflectance, transmittance, size, distance andorientation," said Purves. "This means that there’s no logical way toget back from the retinal image to what’s actually out there in theworld."

Nevertheless, many neurobiologists have attempted to explain visionby postulating that the brain's neural wiring can definitively"calculate" the features of a visual scene, despite the visual world’sinevitable ambiguity. Such “rule-based” theories, said Purves, havearisen because neurobiologists have concentrated on understanding howneurons in the brain’s visual region extract and recognize specificfeatures such as edges in a visual scene.

“Because of the enormouspower and success of modern neurophysiologyand neuroanatomy, there just didn’t seem to be any reason to think muchabout this issue,” said Purves. “However, we began worrying about itseven or eight years ago because the physiology and anatomy people haddescribed didn’t explain what we end up seeing. There was no instance,even in the simplest aspects of vision, where the properties of visualneurons in the brain explain the brightness, colors or forms that weactually see.”

Thus, Purves and his colleagues began exploring visual illusions --the name given to the more obvious discrepancies between the physicalworld and the way people see it -- to understand the strategy the brainuses in perceiving the world. Basically, they statistically comparedperceptions -- such as the apparent length of a line -- with physicalmeasurements of what the line stimulus on the retina was most likely torepresent in the real world.

This sort of analysis, made by measuring a large set of geometricalimages with a device called a laser range scanner, showed that thebrain is not a calculating engine, cranking out stimulus features, buta "statistical engine" wired by evolution and a person's experience tomake the best statistical guess about objects in a visual scene, basedon how successful those guesses have been in the past.

“So, vision is not about extracting features from a scene; it’sabout extracting statistics in the sense of relating the image on yourretina to the visually guided behavior that’s worked in the past,” saidPurves. "This framework for thinking about vision explainsquantitatively -- sometimes in amazing detail -- what we end up seeing."

In 2003, Purves and colleague Beau Lotto published an explanation oftheir “probabilistic” theory of vision in their book "Why We See WhatWe Do: An Empirical Theory of Vision" (Sinauer Associates, Inc).

These two books and dozens of scientific papers have framed thequestions that Purves believes researchers must ask about how visionworks. But he emphasizes that those questions have only begun to beaddressed in neurobiological terms.

"The problem for colleagues in physiology and anatomy is that ourtheory runs counter to what they’ve been doing for the last fiftyyears," said Purves. "And their response has understandably been 'Well,OK, that’s interesting. But how do you relate this concept of vision tophysiology and this anatomy?' It’s perfectly valid to say, 'You’ve gota nice idea and it does explain the phenomenology of what we see, buthow does that relate to the neurons that we know and love?'

"The answer is, we don’t know," said Purves. "That’s going to be thenext many years of vision research. It will mean constructing aframework that explains how neurons and the connections among themoperate in service of this complex, evolved statistical process calledvision.

"Some bright people will certainly do this in the next ten, twentyor thirty years," said Purves. "I don’t expect to be around to see it,but inevitably that will happen. But it’s going to take people whodeeply understand statistics and computer models of neural systems todevelop a working theory of how the properties of neurons andanatomical connections are related to the end product of vision."

Purves said he hopes that the latest book that Catherine Howe and hehave written, along with the earlier work, will continue the process ofenlisting fellow neurobiologists in tackling the immense question ofhow we perceive the confusingly ambiguous visual world around us.