A single-pixel camera that captures complete images by taking many snaps with an array of micro-mirrors could consume less power and produce more compact image files than conventional imaging devices, researchers say.

A conventional digital camera focuses light onto a rectangular array of sensing elements, called pixels, which measure light.

The single-pixel camera developed by researchers Richard Baraniuk and Kevin Kelly at Rice University in Houston, Texas, US, takes a completely different approach. It reflects light from 1024 x 768 micro-mirrors onto a single photodiode. Then it changes the arrangement of micro-mirrors and repeats the process – all in a split second.

The camera switches each mirror randomly between one of two positions – so that they either reflect light onto the pixel or do not. The current version repeats this process about a thousand times in a second, recording the sensor output as it does. A connected computer then works backwards from the sensor output and mirror positions, to generate a complete image.

The device can direct light at the pixel about a thousand times a second. By analyzing data over several seconds it is possible to generate an image comparable to a compressed image from a one-megapixel camera, Kelly told New Scientist.

Terahertz radiation

A single-pixel camera could have advantages over a conventional one, the researchers say. It can record more compact image files and processing the data on an external computer reduces that camera's power requirements.

But the main benefit would be for more exotic imaging devices. Single-element sensors that measure infrared, ultraviolet and terahertz radiation would be far cheaper to make than multi-pixel arrays. The single-pixel system could be used to make cheaper night-vision cameras, for example.

But Kelly adds that single-pixel cameras could help speed up pattern-recognition tasks. Conventional imaging systems produce human-visible images that a computer then analyses. But computers do not need a human-visible image and so could analyse raw data from the camera. The team has already used their camera to recognise that a moving shape is a person. "We're just starting to scratch the surface," he claims.

Kelly and Baraniuk will describe the system at the Frontiers in Optics conference, which opens on 10 October in New York, US.