LAYERED LENS: To take a gigapixel image, which contains more than 1,000 times the amount of information as a megapixel image, in one snapshot requires a special setup. Here, Columbia University researchers propose a ball-shaped lens, half of which is covered by secondary relay lenses, to capture the entire image with minimal distortion. Image: COURTESY OF SHREE NAYAR AND OLIVER COSSAIRT, COLUMBIA UNIVERSITY
More In This Article
Advances in technology tend to spoil us. PCs just a few years old have nothing on today's smart phones, and, whereas megapixel images were once the state of the art in digital photography, gigapixel images (composed of at least one billion pixels, or picture elements) are beginning to show up on the Web in vivid detail.
Gigapixel images also hold tremendous potential for providing law enforcement and the military with detailed reconnaissance and surveillance information. Long-distance images taken today by satellites or unmanned aerial vehicles (UAVs) can capture detail down to a license plate number while flying at altitudes too high for these drones to be spotted from the ground. But these images provide only a narrow view, says Ravi Athale, a consultant to the Defense Advanced Research Projects Agency (DARPA) and a senior principal scientist at MITRE Corp. in McLean, Va. He likens UAV images to seeing a battlefield or city through a "soda straw" and satellite images to an injection needle.
"We are no longer dealing with fixed installations or army tank units or missile silo units,” Athale says. “[Fighting terrorism requires] an awareness of what's going on in a wide area the size of a medium city."
Through its Advanced Wide Field of View Architectures for Image Reconstruction and Exploitation program, DARPA has for the past year been working on ways to develop a camera that can take a gigapixel-quality image in a single snapshot. This approach is novel, given that today's gigapixel images actually consist of several megapixel-sized images pieced together digitally to provide a high level of detail over a large area. This is often done using a long-lens digital single-lens reflex (SLR) camera placed atop a motorized mount. Software controls the movement of the camera, which captures a mosaic of hundreds or even thousands of images that, when placed together, create a single, high-resolution scene that maintains its clarity even when the viewer zooms in on a specific area. DARPA plans to invest $25 million over a three-and-a-half-year period in its program, which includes a component called Maximally scalable Optical Sensor Array Imaging with Computation (MOSAIC).
The single-snapshot approach to gigapixel digital photography has its drawbacks. The equipment is bulky, expensive and complicated. In addition, because it may take several minutes or even hours for the automated camera to shoot all of the individual images required to create the larger mosaic, lighting conditions may change and objects (cars, people, aircraft, etc.) can move into and out of the frames. And stitching together the individual images requires software that must match overlapping points—any errors must be corrected manually.
Such images also require special viewing software found on Google Earth, 360world.eu, Gigapan.org (created by Pittsburgh's Carnegie Mellon University, NASA and Google) and other Web sites that allow gigapixel digital photographs to be uploaded, viewed and shared across the Web.
Nor are gigapixel images conducive to being captured by a compact, inexpensive camera. The digital processors and memory used in today's cameras are ill-equipped to manage gigapixel images, which contain more than 1,000 times the amount of information as megapixel images. (A 10-gigapixel image would take up more than 30 gigabytes of hard drive space.) And although pixels are often used in reference to image resolution, this attribute can truly only be measured by taking into account an image's overall dimensions and the number of pixels per inch or per centimeter. For example, an image that is 20.3 by 25.4 centimeters at 60 pixels per centimeter has the same resolution as an image that is 10.2 by 12.7 centimeters at 120 pixels per centimeter.
A team of Columbia University researchers in New York led by computer science professor Shree Nayar thinks a single snapshot gigapixel camera is possible if they can reduce the complexity of such images. "Rather than thinking about it as capturing the final image, you're capturing the information you would need to compute the final image," Nayar says.
In a paper to be presented at the April IEEE International Conference on Computational Photography (ICCP) in Pittsburgh, the Columbia researchers propose three relatively compact camera designs (two of which they have actually built as prototypes) for single-shot gigapixel imaging—each design relies on a ball-shaped lens and one or more digital sensors. Such a lens is one of the simplest because it has perfect symmetry (leading to fewer aberrations) and consists of one element rather than several lenses that must be configured to work together, says Oliver Cossairt, a Columbia computer science PhD candidate who works with Nayar.