A new single-pixel camera might soon outperform current digital cameras that rely on millions of pixels to create a high-resolution image.
Credit: iStockphoto
SYDNEY: A new digital camera that captures high-resolution images using a single sensor might soon outpeform current technologies that rely on millions of sensors, U.S. researchers say.
"We've developed a new single pixel camera that from just a few random measurements can reconstruct a high-resolution image," said Richard Baraiuk, one of two engineers from Rice University in Houston who developed the technology. The camera will be unveiled at an Optical Society of America conference in New York this week.
Compared with conventional digital cameras, this new technology requires less power, takes up less room and captures light from previously undetected parts of the spectrum.
But don't rush down to your camera store just yet. At the moment, there’s only one single-pixel camera in the world. It’s sitting in the corner of a laboratory at Rice University and currently takes about five minutes to take a single picture … of a stationary object.
In the future, however, the researchers hope to capture images with the speed of current digital cameras. They predict their technology will first be developed to take pictures in the X-ray, infrared, and ultraviolet regions of the spectrum, as sensors in these regions are much more expensive – often worth more than A$1,000 each. Reducing the number of sensors required from millions to just one could make photography in these spectral regions affordable.
But for now, Baraiuk and his colleagues are satisfied with proving that a single-pixel camera that captures high-resolution images is possible.
"We wanted to attack some of the inherent inefficiencies in conventional digital photography,” said Baraiuk. “In particular, the fact that we sample digital images with millions of pixels but then need to compress the resulting data, essentially throwing away much of the data."
A digital camera works by turning light into a numerical sequence. In a five megapixel camera there are five million sensors, or pixels, so a single image is converted to five million numbers.
This is far too much data to store, so the camera approximates the picture by compressing the numbers. Just like the word Australia can be compressed to Aus (or Oz) without changing its meaning, the numbers representing an image can be compressed without losing much detail.
This process is fairly wasteful, according to Baraiuk. Five million pixels are built, while only 50,000 to 200,000 numbers are used to represent the original image. And this number-crunching process is power-hungry, draining the camera's battery.
Like conventional cameras, this new camera uses a lens to focus the light. However while light strikes five million sensors in a conventional camera, in this new camera the light hits an array of tiny mirrors.
Each mirror can be controlled to either bounce light into the single sensor, or away from the sensor. The array of mirrors goes through a sequence of random patterns, producing a 'random measurement'.
The camera takes about 200,000 random measurements with the single sensor. Using some funky mathematical tricks the original image can be extrapolated from these random measurements.
The most bizarre part is that the patterns have to be random, it will not work as well if a regular pattern is used. Essentially, this random measurement is just white noise, similar to the fuzzy screen you get on a television when you tune into a dead channel.
"White noise is the key," according to Baraniuk. "Thanks to some deep new mathematics developed just a couple of years ago, we're able to get a useful, coherent image out of the randomly scattered measurements."
Single-pixel camera
A telescope made of a long drooping reflective wire, turned about the axis so that the resulting reflected image is the night sky above...It looks like a radar sweep to a camera and is somewhat averageable to my computer. But is sounds like you fave a better math program. Although I can make out stars and record an image for averaging; crashing a lot has been my fate. Maybe you'll do better than I did with my math. Give it a go.
Sounds like one of the
Sounds like one of the biggest advancements for digital cameras in the last years. And if this new technology can improve x-ray machines and such maybe it can do even more things. I look forward to hearing more about this research and one day having my own model.