Single-pixel has multiple futures

Single-pixel has multiple futures
Terahertz version adds boundless potential to unique Rice-built camera

BY MIKE WILLIAMS
Rice News staff

A few years ago, Rice professors Kevin Kelly and Richard Baraniuk made a camera that had something in common with the original, and yet was not like any camera ever made.

”We used to have a joke at the end of our talks,” said Kelly, an associate professor in electrical and computer engineering. ”We’d show the first photograph (by Frenchman Joseph Nicéphore Niépce in 1826). He made a plate of reactive chemicals and used it to capture a scene outside his window — but it was an eight-hour exposure. And at the time we were giving those talks, it would take us about eight hours to reconstruct an image from our camera.

”These days, we can do it in about 10 seconds with a regular laptop.”

	JEFF FITLOW
	Kevin Kelly, associate professor in electrical and computer engineering, with elements of the single-pixel camera at the Abercrombie Engineering Laboratory.

Such is the state of Rice’s now-famous single-pixel camera, a technology that made waves when it was introduced two years ago, supported in part by the National Science Foundation and NASA. Advances have come quickly, particularly in the compressive sensing algorithms that make it possible to do with one pixel what takes commercial digital cameras millions.

You can learn the basics here, but essentially, the original camera uses a digital micromirror device (DMD) of the sort found in projection televisions to send a rapid-fire series of randomly sampled pieces of an image to the pixel sensor, or photodiode. Those thousands of samples are then reassembled into an image, a math-intensive process that falls into the domain of Baraniuk, the Victor E. Cameron Professor in Electrical and Computer Engineering; assistant professor Wotao Yin; and his mentor, professor Yin Zhang, of the Computational and Applied Mathematics Department.

The pocket version of the single-pixel camera may never come to a Wal-Mart near you — the original and its siblings exist only as strings of little black boxes and data-crunching computers in Kelly’s lab. But the technology’s potential seems unlimited for applications that need to gather a lot of data and don’t have the computer horsepower to crunch it on-site. Think, for example, of satellite-based systems like GPS or telescopes.

”Any kind of application that requires the transmission of sparse data streams will benefit from this,” said Yin, who came to Rice two years ago and was awarded a National Science Foundation CAREER grant earlier this year to continue his research on compressed sensing.

Scenarios in which only a small amount of data is required to make a fast decision could benefit. ”A GPS is sending you a lot of information when, really, all you need to do is decide whether to go right or left,” he said. Spending unnecessary time and computational energy on a problem is wasteful, so providing just enough of the correct data is the way of the future.

That will be important as compressive sensing evolves from a system that handles snapshots to one that handles streams of information for, say, video-based applications. ”That may keep me here working on it for 20 years, but that’s OK,” said Yin.

The advances ”could make for very inexpensive security and scientific cameras in the near future,” said Baraniuk. He, Kelly and Rice Professor Daniel Mittleman, graduate students Wai Lam Chan and Dharmpal Takhar and undergraduate student Kriti Charan published a paper in Applied Physics Letters describing one such possibility, a terahertz imaging system that has potential applications in security, telecommunications, signal processing and medicine.

Terahertz radiation, which occupies space in the electromagnetic spectrum between infrared and microwave, penetrates fabric, wood, plastic and even clouds, but not metal or water. As their rays are not harmful, cheap terahertz cameras may someday be used for security screening in airports, supplementing traditional X-ray scanners and walk-through portals.

”There are lots of applications for terahertz imaging, if you could make a real-time imager that’s sensitive enough. Some of them are pretty science-fictiony, but some are pretty realistic,” said Mittleman, a professor in electrical and computer engineering. ”I think this is really promising.”

This illustration shows components of the experimental Terahertz camera, based on the single-pixel camera developed by a team of Rice researchers.

The terahertz camera ”was a little harder to implement than we thought,” said Kelly, describing their research as a proof-of-concept project. For this camera, they replaced the DMD with a series of copper sheets — 600 of them — that allowed terahertz radiation to reach the pixel through random pinholes.

”There’s very good reason to believe you could build a terahertz modulator that could do that same task electrically and very fast,” said Mittleman. Los Alamos National Laboratory has sent just such a modulator to Rice for testing in Mittleman’s lab; he described the device as a four-by-four array of metamaterials that become opaque to terahertz radiation when a voltage is applied.

”If that works, we’re going to go up to a 32-by-32, and at that point, you’ve got enough to make images,” he said. ”What we really want is 1,000-by-1,000, but the fabrication challenge there will be substantial.”

The hyperspectral capabilities inherent in even a basic single-pixel camera make it useful for all kinds of things, said Kelly. ”Current cameras break an image down into red, green and blue,” said Kelly. ”But this system breaks down every pixel into all the individual wavelengths that make up a color.

”If you want to know whether that green object over there is a bunch of trees or a tank painted green, this system will tell you,” he said.

Such super spectrometers could find homes in satellites pointed at Earth and at the stars. Instead of putting power-hungry image processors on spacecraft, he said, the cameras themselves would send raw data back to Earth for processing on the ground — a much faster, cheaper solution.

Inner space could benefit, too. ”To do stuff beyond visible light with a confocal microscope, the sensor becomes amazingly expensive,” Kelly said. ”For example, to go from visible to infrared, you go from a $100 CCD (charge-coupled device) array to a $50,000 CCD array. With ours, you go from a $1 photo diode to a $10 photo diode. That’s it.”

Kelly said the team is now working to replace the micromirror device, the camera’s most expensive component. ”Instead of the DMD, you can take a piece of paper that’s covered in something shiny and flex it in lots of different, random ways. As long as you know how you’re flexing it every time, and as long as it’s repeatable, this would work equally fine,” he said.

”One of those Mylar birthday balloons could be your mirror. Then, suddenly, a single-pixel camera becomes $10 instead of $500. That’s the next evolution.”