Donald Bren School of Information and Computer Sciences

I surveyed work on visual object recognition and perception. In animals, vision has been studied mainly on the behavioral and neurobiological levels. Behavioral data typically show what the visual system, by itself or together with the rest of the organism, is capable of. They show, for example, that humans can recognie objects regardless of size and position, but that rotated objects pose problems. Important insights into the organization of behavior have also been provided by people who suffered localized brain damage. We have learned that the brain is divided into areas subserving different and relatively well-defined behaviors. The visual system itself is also organized in different subsystems; the visual cortex alone contains nearly twenty maps of the visual field. And individual neurons respond selectively to visual stimuli, e.g., the orientation of line segments, color, direction of motion, and, most intriguingly, faces. The question is how the actions of all these neurons produce the behavior we observe. How do neurons represent the shape of objects such that they can be recognized? Before we can answer the question, we have to understand the computational aspect of shape representation, the nature of the problem as it were. Many methods for representing shape have been explored, mainly by computer scientists, but so far no satisfactory answers have been found.