New slant on vision research: Neurons sensitive to viewing angle

Mistakes made by human subjects in identifying the facing direction of faces, cars or meaningless objects have yielded evidence that the brain contains nerve cells, or neurons, whose job is to encode the viewing angle of objects. It is well known that certain neurons respond to color, motion, edges and other aspects of our environment. Now, University of Minnesota researchers have found that our visual cortex contains neurons that tell us, for example, whether a face is turned in our direction or not. The work adds to knowledge of how the brain collects and processes visual information leading to the recognition of objects, and it may inform the design of machine vision. The study will be published in the March 3 issue of the journal Neuron.

The brain relies on millions of neurons to report the visual elements of our environment. But, for example, if every neuron geared to motion fired in response to any motion whatsoever, then we couldn’t tell whether a train was chugging into the distance or bearing down on us. Instead, to gain a complete picture of the world, our brains appear to contain separate, but physically intertwined, populations of neurons that respond to only one small aspect of our environment. The brain then bases its interpretation of images largely on which neurons fire.

“The issue is, what is the underlying neural mechanism that supports the ability to recognize objects viewed from different angles?” said Sheng He, associate professor of psychology, who directed the study. “This study supports the idea that we have explicit representations in our brains for specific views of objects.” The study was carried out jointly with Fang Fang, a graduate student in He’s laboratory.

The researchers presented volunteers with the image of a face, a car or a meaningless geometric object on the computer. In each case, the first image–called the adapting image–was turned to one side. After a very brief pause, another image of the same face or object–called the test image–was flashed on the screen. But this time, the image was either head-on or turned very slightly (three or six degrees) to one side or the other. Whatever the orientation of the test image, subjects were required to choose whether it was turned to the right or the left.

When subjects were presented with an adapting image turned 30 degrees to one side, then tested with an image of the same thing in head-on view, they tended to say the test image was tilted in the opposite direction of the adapting image. That is, if they first saw the face of a man turned 30 degrees to the left, then saw his face head-on, they said the face was turned to the right. This “adaptation effect” occurred 80 percent of the time; normally, responses for both directions would be equally likely. Even if the test image was turned three degrees in the same direction as the adapting image, the subjects guessed wrong half the time, saying the test image was turned in the other direction.

The reason for the errors is that when a person stares at an image, neurons that respond to the viewing angle of the image get “tired” and become less responsive when a very similar image is presented again, He said. The brain interprets this lack of response as the object not being turned in the direction the neurons are attuned to. This suggests that there are separate populations of neurons, each responding to a particular narrow range of orientations. The neurons are likely located in the lateral occipital cortex, an area of the cerebral cortex very far back on either side of the head.

The researchers also performed experiments that suggested that for faces, at least, subjects were not deciding the orientation of test images based on “local” features such as noses. When subjects saw unorganized fragments of faces–as if parts of the face were simply erased–as adapting images, no adaptation effect occurred.

“This shows that fragmented local features are not sufficient to get the adaptation effect,” He said. “You must have a global representation of the face. But local features may be important within the context of a [complete] face.

“Also, if you adapt to a face, then test with an image of a car, you don’t get adaptation. So local features that identify the object as a face or a car are important. Researchers think there are populations of neurons that respond to classes of objects in the environment, for example, houses, hills, tools, faces and so forth.”

The researchers next plan to put volunteers in a functional magnetic resonance imaging scanner and see how different neurons respond to different views of objects. The work was supported by the James S. McDonnell Foundation, the National Institutes of Health and the University of Minnesota.