Next: A View from Neuroscience
Up: Neural Coding in the
Previous: Contents
  Contents
Humans perceive visual input amazingly well. For much of the history
of artificial intelligence, researchers have tried to make a computer
match the perceptual capabilities of a human being, and the results
have not been very impressive, because the task is so difficult. To
appreciate the difficulty, consider looking at a person's face. As
humans, we can see a person's face for a split second and know who
they are and often how they feel at that moment. But the computer
sees only a collection of millions of pixels of different colors. A
huge number of those pixels are different between two pictures of a
person if the person moved slightly to the left in one picture, or if
the person is a little farther away, and therefore smaller in the
image. One could imagine for such cases that the computer could
figure out which pixels moved to which pixels in the image, since they
are the same color and same relative position. But what if the person
turned their head a little? Then many of the pixels in one image are
not even in the other. Or what if the lighting is different? Or they
changed their expression? Or their hair moved? All of these cause
huge variations in the image, and we as humans barely perceive them at
all. Instead, we perceive only what is interesting and useful to us.
How people tune out what is uninteresting and see only what is useful
has been the focus of a great amount of research over the years.
There are a number of reasons this problem is of such great interest.
Most obviously, if we could understand human vision, we might be able
to make computers do the same thing. If computer vision were on par
with human vision, the applications for it would be endless. But the
goal is more broad than that. There is evidence to indicate that what
the brain does in the visual system is also done in the auditory
system. If we could figure out vision, we would probably understand
hearing as well, or at least be much closer to it. But if the same
thing is done in both vision and audition, then it may be done
elsewhere. The same principles which the brain uses to process vision
may be instrumental in our understanding of reasoning. The visual
system may serve as the Rosetta stone for the rest of the brain.
Next: A View from Neuroscience
Up: Neural Coding in the
Previous: Contents
  Contents
Jonathan Shaw
2003-09-30