Digital noise sharpens vague images - visual processing research

Science News, March 30, 1996 by Richard Lipkin

A good, clear picture may be worth a thousand words. But how much is a fuzzy image worth?

That depends on how much information a viewer can obtain from the image and on whether the information proves useful. Enrico Simonotto, a physicist at the University of Genoa in Italy, and his colleagues have found that adding randomized signals or backgr ound noise resembling the snow seen in weak television pictures sometimes enhances a faded image.

Adding noise, it seems, can lift a barely detectable image above the brain's perceptual threshold so that people viewing the image can grasp some details that would otherwise be lost.

"Our goal is to see how noise affects the way the brain processes information," Simonotto said at last week's meeting of the American Physical Society in St. Louis.

Starting with a clear digitized image of a face, the researchers used computer graphics to lower the contrast until the features were no longer distinguishable. The group then added randomized digital signals, which can be described mathematically as a ty pe of stochastic resonance (SN: 7/22/95, p. 55).

"We found that by adding noise alone, some of the original picture's details could be perceived," Simonotto says. Moreover, by testing noise at different frequencies, the team further improved the quality of the picture.

The brain somehow uses the noise to reconstruct pieces of the picture lost from the original. "If you look at a weak image of my face, for example, you get a hint of a human face but not much more," Simonotto says. "Am I wearing glasses? Without adding no ise to the image, you can't tell."

The amount and type of noise added to the image affects the way viewers discern a picture's details. For example, a fast, fluctuating noise enhanced images more effectively than static noise did.

Theories of stochastic resonance arose in 1981 as physicists sought to explain the periodicity of Earth's ice ages. Subsequently, scientists brought the mathematical theory to bear on biological problems, using it to describe how animals such as crayfish sense their environment. Meanwhile, neuroscientists were also learning that the brain, despite its exquisite precision as an information processor, generates much internal noise, says Frank Moss, a biophysicist at the University of Missouri-St. Louis.

"Neurons are noisy," Moss says. "If you measure signals in the brain or in a sensory organ, you mostly detect random firings. One of the brain's strengths as a computer is its ability to extract information from noisy signals."

"Think of the brain as an instrument filled with sloppy amplifiers," says Martin B. Stemmler, a computational neuroscientist at the California Institute of Technology in Pasadena. "Our visual system averages lots of signals, distinguishing real ones in th e external world from internally generated noise of the brain's own circuitry. This is all part of successful image processing.

"Simonotto's work brings together knowledge from video engineering, computational neuroscience, and the theory of stochastic resonance," Stemmler says. "An interesting question to pursue is how the brain uses noise to enhance images."

Though this work remains preliminary, Simonotto says it may someday prove useful in systems that help humans see in visually challenging circumstances-at night, in snow or fog, or underwater.