Picture This: Imaging Technology That Makes Machines ‘See’

The lightning-fast evolution of the camera is leaving our old Brownies in the dust. The digital camera of the distant future may produce three-dimensional images, and it might not have a lens.

A team of researchers at the University of Illinois, Champaign, has taken a step in that direction with the creation of an imaging system that doesn’t use a lens. The advantage is an infinite depth of field, so everything in view of the camera remains constantly in focus.

A computer does the work of the lens by combining many two-dimensional images into a three-dimensional picture.

This is cutting-edge research, and it is a long way from the marketplace. But it points the way to new systems that could eventually give vision to robots and allow researchers to create 3-D images of cell-sized structures.

The technology weds the mathematical foundations of two very different areas of imaging: interferometers, used by radio astronomers, in which one radio wave from an object is superimposed on another, yielding very precise measurements of position and intensity; and a form of X-ray photography called cone-beam tomography.

The system works by taking a series of “snapshots” of an object bathed in ordinary light as it is rotated in front of the camera. Each snapshot consists of a distinct slice of the object through a given plane as the light reflects back to the camera.

“The thing that’s unique about those pictures is that all of the planes are in focus on the imaging screen because there’s no dispersion, which would cause the object to go out of focus like with a conventional lens,” said Ronald Stack, a research engineer with the university’s Beckman Institute for Advanced Science and Technology.

Daniel Marks, a graduate student, recognized that by applying a mathematical formula known as a Fourier transform, the “snapshots” could be recombined into a 3-D image.

The researchers feed the data into a personal computer, which rebuilds the image one slice at a time, then provides the three-dimensional picture. At this point, it’s a slow process, and it’s a memory hog.

“When we do the reconstruction, it does require quite a bit of memory,” Stack said. “If we do a cube [512 pixels on each side], then we will need somewhere on the order of 512 megabytes of memory. That’s not completely unreasonable for computers these days.”

But the reconstruction may take as long as six hours, making the current system impractical for real-time imaging. All of those factors could be improved, Stack said, as the researchers refine algorithms used to reconstruct the image.

The researchers discussed their “lens-less camera” in the June 25 issue of the journal Science, but they have already launched another project that could produce instantaneous 3-D images. They are using 64 digital cameras that can be focused individually on a target. The cameras have lenses, so each must be focused on a specific area of the object.

The data are fed from the cameras into a supercomputer, which reconstructs the three-dimensional image in real time. Eventually, that system is to be linked to a virtual reality “cave” on the university’s campus.

Better imaging technology is critical for scientists struggling to reach the very elusive goal of building machines that can “see.” The human eye is amazing in its ability to distinguish incredibly subtle differences in the visible world. If robots are ever to play key roles in such different arenas as medical surgery and space exploration, vast improvement must be made. And there is some progress.

Researchers at Pennsylvania State University disclosed that they are developing a computer vision system that tracks moving targets.

Octavia I. Camps, associate professor of electrical engineering and computer engineering, said their system works by programming the computer to recognize parts of an object, not the object as a whole.

“Instead of learning to identify a whole object, the computer can be trained to identify an object by its significant parts,” Camps said. “So if another object is in front of it or the camera is out of focus, the computer can still recognize it.”

That addresses one critical component of the “machine vision” problem: how to design a system that can unfailingly recognize something that looks a lot like something else, but is different.

Even the human eye doesn’t always do that.

Lee Dye can be reached at leedye@ptialaska.net.