A normal digital camera focuses an image onto a grid of pixels, each of which measures the intensity of light at a spot in the image. The more pixels, the sharper those pictures from your last vacation turn out. But denser pixel arrays also drain a camera's battery faster, and most of the information they record ends up unused. To store images, the camera spends more power to convert them into a file format such as jpeg, which removes redundant information and fine details.
Aiming to streamline digital imaging, Rice University engineers Rich Baraniuk and Kevin Kelly replace the pixel grid with an array of tiny micromirrors. Each mirror points in one of two directions at random: either toward a single light-sensitive pixel or away from it. When an image is focused onto the mirrors, the light breaks into a random pattern of bright spots, and their combined intensity is recorded by the single pixel. "What this gives us is simultaneous acquisition and compression of an image," says Kelly. The positions of the mirrors are randomly reassigned up to tens or hundreds of thousands of times, creating a series of randomly varying light intensities. Using a computer algorithm developed by mathematicians over the past few years, the camera software identifies the simplest possible image that is consistent with the samples, essentially decompressing the image. "It's kind of a heroic demonstration of the far limits of this concept," says engineer David Brady of Duke University, who works on similar techniques for designing thin- and wide-spectrum cameras.
The Rice group plans to report at the Optical Society of America's annual meeting on October 11 that they have created megapixel-size images based on 30,000 light samples obtained in about 15 minutes. Faster switching of the mirrors could in principle reduce that time to seconds, Kelly says. "Initially it'll be a research tool because cameras are very inexpensive," he notes. "But the same method could be applied without using the mirrors." By playing with light captured in conventional cameras, he says, the algorithm could produce sharper images from the same number of pixels.