MIT develops new Fourier algorithm for image tech

MIT researchers have put forward a paper for a new approach to fast Fourier transform math that could provide a major lift to image compression and other signal processing technology. The new technique, discovered by associate professor Dina Katabi and professor Piotr Indyk, divides signals and looks for "sparse" but strong frequencies within each section of the signal. Since it would only need to sample random details from those sparse signals instead of full details, it could speed up the processing time by as much as ten times, MIT said.

One version of the technique would borrow from 4G cellular data processing to find dominant frequencies and quickly find what they need. It would also merge filters during the process to avoid signals that would be too attenuated at the edge of a given filter and thus couldn't be sampled otherwise.

Earlier approaches tried speeding things up, but often broke down if there were too many of these important signals to sample relative to the total number of signals. This approach degrades, MIT said, but unlike before still improves on earlier fast Fourier transforms.

The speed-up could allow for much tighter photo and video compression that could have particular benefits in mobile. High-resolution images or long movies could consume less space and consequently would use less bandwidth and battery life to send. Audio files could also get smaller, and certain kinds of general math would also accelerate.

How soon the new algorithms could reach shipping software and products wasn't mentioned. As math rather than a physical product, though, it could arrive faster than other forms of research. [via MIT]

An example of dividing a signal into sections to process the total result

By Electronista Staff