Peter Gerhardstein is relying on infants to help him unravel one of the most perplexing of developmental mysteries: how is it that humans come to recognize the objects they see in the world around them.
While most of us unconsciously accomplish this task hundreds of thousands of times a day, object recognition is something that computer vision researchers, adult perception psychologists and a host of other researchers have been attempting to understand for more than 40 years. For all their best efforts, researchers have made remarkably little progress, Gerhardstein said.
People have this ability to see and to recognize objects in the world. In general, we don’t even think about this because our vision system is so good at it. It turns out, though, that this is actually a horrendously difficult thing to do. The ability, for example, to see a chair from all different orientations and to understand and know that it is the same chair and this is terribly difficult.
An assistant professor of psychology, Gerhardstein joined the University two years ago after arriving as a research assistant in 1997. This fall he received a five-year $588,961 grant from the National Institute of Child Health and Human Development for his research project Viewpoint Invariance in Infants Perception of Objects. Working with three- and six-month-old infants, Gerhardstein hopes to help discover how it is that humans develop the ability to accomplish this surprisingly complex task a task that clearly relies not just on sight, but also on memory.
By studying vision at various developmental stages, Gerhardstein hopes to understand the developmental path of object recognition, which has already been closely studied in adults.
We have a pretty good understanding of how good adults really are at this, but we don’t really know how they get to be that good or exactly what they’re doing, he said.
The most rudimentary of theories suggests that people become capable of recognizing an object when they see it in enough detail to form a picture of it in their head. That simplistic theory suggests that the next time that person sees the same object or a similar object, no matter the angle, they can perform a mental rotation of the picture and recognize the object.
Gerhardstein says there is strong evidence to suggest that this simple explanation can’t really explain the complete process of object recognition. Studies suggest, for instance, that adults are generally capable of recognizing objects more quickly than would seem plausible if they had to rely on their brains to figuratively pick up and examine a stored representation of an object from all angles.
Instead, Gerhardstein thinks a theory developed by Irving Biderman, his doctoral advisor and the William K. Keck Professor of Cognitive Neuroscience at the University of Southern California, is more informative about the basics of what might really be going on in the process of object recognition.
Biederman’s theory posits that objects and scenes are represented in the human brain as an arrangement of simple, viewpoint-invariant shapes that Biederman terms geons. Geons include such shapes as bricks, cylinders, wedges and cones. They are called viewpoint-invariant objects because, as a general rule, they are perceptually recognizable when viewed from the top, bottom or side the viewpoint doesn’t really matter.
Using Biederman’s theory as a starting point, Gerhardstein and his research team began working with three-month old infants to tease apart and examine the development of object recognition abilities.
During the initial phase of Gerhardstein’s project, three-month old infants were taught to kick a mobile from which brick-shaped objects were suspended. The reward for this behavior is that when the mobile is kicked the shapes move and a compact disc plays a Sesame Street song. Because babies love any activity in which they gain some control over their environment, they generally take to this activity easily and will enthusiastically engage in it for long periods of time, Gerhardstein said.
The babies learn this game for two days, during which time their kicking rate tend to go up as they figure out that they can control the music and the movement of the brick-shaped objects by kicking the mobile. The babies are then allowed 24 hours down time. On the third day of observation, except in the control group of infants who were allowed to go on merrily kicking at the same mobile, the brick-shaped mobile is exchanged for a cylinder-shaped mobile. If the babies are unable to discriminate between the two shapes, they can be expected to keep kicking at the pre-exchange rate. If they stopped kicking, it could be presumed that they recognized the cylinders as being different from the bricks.
Remarkably, what Gerhardstein’s preliminary findings suggest is that even three-month-old babies are able to discriminate between bricks and cylinders. Their kicking comes to a quick halt when the bricks are replaced with cylinders. But perhaps even more revealing, a separate portion of the study suggests that the babies don’t react differently if only their view of the bricks is altered.
According to the infants, he said, a different viewpoint doesn’t matter. What this is suggesting is that if these geons or parts are the basis of the way we construct object representations, infants may indeed already possess this knowledge. They might not need to learn these parts; they might start out in life with a sort of visual alphabet.
This is the first test, however, and the infants are allowed a lot of time to learn. With less learning, they might not show this ability, Gerhardstein added.
In any case, strong evidence suggests that adults somehow end up with the viewpoint invariant object-based representational system that Biederman has described, meaning that they are generally able to recognize geons like a brick as the same object from any angle. Gerhardstein’s research could be helpful in beginning to identify how object recognition abilities develop from infancy to adulthood. These geons may be the basic building blocks from which infants develop object recognition.
As a general rule, more and more developmental research suggests that it is the interaction of genetics and environment that makes things happen, Gerhardstein said. He wont be surprised, therefore, if it turns out that a synergistic interplay of experience and physiological maturation prove key to a fuller explanation of the phenomenon of object recognition.
This is an attempt to describe how the development of the vision system normally progresses so that we might then start to make conjectures about what exactly is happening when something goes awry, he said of his current project.
That’s why the National Institute of Child Health and Human Development is so interested in his work.
If we are going to come up with a reasonable explanation for things that go wrong during development, we need first to understand the normal course of development.
It will also be extremely useful if we can divide things into this just matures versus this is probably impacted by experience. We can then start to understand better how we might best go about treating things that go wrong.
There would be little point, for example, to the use of cognitive behavioral therapy if the ability to recognize objects is something that either matures or fails to mature based only on biological or physiological factors.
In that case you are much more likely looking at a pharmacological treatment or you may in fact be looking even earlier at gene screening, Gerhardstein said.
Gerhardstein said that if his preliminary findings hold up, they could help flesh out Beadsman’s theory and support the theories of Albert Yonas, a noted child development professor at the University of Minnesota. Yonas work centers on how infants and preschool children come to perceive the visual world. His work isolates the visual information that makes perception possible and explores when infants become sensitive to that information. Partly as a result of his work, it’s clear that infants don’t develop stereovision until about four months of age. They are also unable to understand the implied depth of drawings, Gerhardstein said.
All of this points to the notion that infants are working with an impoverished world to begin with. We very much want to know what kind of consequences this sort of thing has for them in object representation and recognition.