Summary: Visual Information Processing

Visual scene analysis is a computer-based effort to simulate visual perception. It requires the computer to identify which parts of a scene belong together to form objects.

A key part of the process is constraint satisfaction. The computer locates edges and corners where lines come together (called vertexes). The assignment of meaning to each line and vertex has the effect of limiting or constraining the interpretation put on other parts of the scene.

The constraints propagate. Each part of the scene, when interpreted, helps to limit or suggest interpretations for other parts of the scene. Constraint-propagation eventually produces one interpretation of the whole scene consistent with all the evidence, much like a detective finding a crime solution that fits all the clues.

Classic work on visual scene analysis in the mid-1960s culminated in a successful program by the 1980s. Success, in this case, was defined by a computer's ability to start with a visual image from a camera and identify which lines, edges, and corners belonged to particular objects. The ability to do this quickly and accurately is crucial to the success of self-driving cars.

Humans and computers both use a combination of bottom-up (data driven) and top-down (schema-driven) processing. Bottom-up processing is the flow of raw data into the system that helps determine activity at higher levels.

Top-down processing is the influence of pre-existing knowledge used to interpret incoming data. A schema is a pattern or piece of knowledge based on past experience.

The phenomena of set and expectancy were illustrated by the cartoon-like rat man example. People were swayed toward one interpretation or the other by a series of images leading up to the rat-man image.

Cartoonists exploit top-down processing. They use tiny cues (such as beads of sweat to show effort) that readers have learned to associate with emotions, movements, and mental states.

Mental imagery takes place in imagination. Cooper and Shepard showed that the speed of rotating an imagined shape varied with the amount of rotation required. This suggested that spatial-type mental images had characteristics similar to physical models that could be manipulated.

Kosslyn showed the same thing for picture-like mental images. People treated them almost like models that could be scanned.

Brain areas involved in normal perception are also involved in mental imagery. However, picture memory and spatial memory are distinct, involving different areas of the brain.

After time passes, memory of images is based on the story told by images, rather than exact visual details. Baggett's "haircut" experiment showed that people lost memory for details of an image within three days.

Prev page | Page top | Chapter Contents | Next page

Write to Dr. Dewey at psywww@gmail.com.

Don't see what you need? Psych Web has over 1,000 pages, so it may be elsewhere on the site. Do a site-specific Google search using the box below.