Machines kick our fleshy little butts at many activities: They're stronger lifters, faster multipliers and better chess players. But with all of the computer's talents, there's still one mechanical function we can perform that computers cannot: We can gaze upon a collection of lines and perceive objects.
In a paper published in the January 5th issue of Neuron, Johns Hopkins researchers Scott Brincat and Charles Connor present evidence for a theory of how we see integrate visual elements into complete objects. Apparently, we start by recognizing individual object parts—bits of contour—and gradually piece the puzzle together, recognizing groups of features, and, eventually, full shapes. The whole process of grouping and assembling takes about 60 milliseconds.
Advertisement
"It may represent a basic mechanism for putting together simple information to derive more complex information about the complex entities in our world," said Connor.
According to Connor, after being presented with a visual stimulus, cells first signal information about local parts of the shape, not the shape as a whole. Each cell is tuned into a specific feature of contour—such as orientation, curvature and location—relative to the rest of the object; if the shape has that feature, the neuron fires rapidly.
"A given cell will actually respond to lots of very different shapes, but the common characteristic will be something like, all of those shapes have convex curvature near the top," Connor said. "So that cell is signaling something about convex curvature near the top of the shape rather than global shape."
In this study, the researchers examined cells in the stage following first glance: information synthesis. They recorded how rapidly individual cells fired when subjects viewed a set of stimulus shapes with subtle variations in geometric values. The researchers then examined which cells fired rapidly in response to which shapes.
Connor said cells involved in the synthesis phase—which are located in a different part of the brain from those that recognize fragments—respond to groups of several contour regions located in the same shape. The first step of synthesis yields an ambiguous signal: A cell that responds to a certain group of, say, three features will fire if any one, two or all three are present. Over the course of the next 60 milliseconds, however, this signal is refined, and a cell that responds to three features will only fire if all three are present. The brain thus goes from noticing small contour fragments to recognizing larger configurations.
"That dynamic process of refining the representation," Connor said, "gives you an object representation that's much more useful, more compact, more closely related to the kind of complex shape configurations that we need to know about in our world."
Connor believes this study brings science one step closer to understanding the incredibly complex process of human vision. He noted that the brain has to be able to discern an object, even as the visual input varies due to changes in position, size and angle of the object in the retina.
"So what the brain is looking at, coming from the retina, is like a million entry long list of pixel values in a display screen," he said. "You and I couldn't just read through one of those lists and figure out what was being displayed on the screen; you'd have to do lots of computations to figure out what was happening.
"And that's basically what these visual parts of the brain are doing."