Algorithm Creates High-Resolution Images from Low-Resolution Pictures

February 1, 2011

Technologies using sensors that produce only low-resolution images may now be able to produce higher-quality images without needing additional computing power or a bigger footprint, thanks to software being developed by an MDA-funded company.

Numerica Corporation (Ft. Collins, CO) has developed algorithms that take a series of low-resolution images (like the one shown in the figure) and process them to form a higher-resolution picture.

This innovation mostly revolved around target-tracking problems. MDA gave the small company, which works in the applied math field, a $750,000 Phase II SBIR contract in 2006 to enhance the agency’s ability to intercept missiles in mid-course.

Boosting infrared (IR) sensor resolution presents some challenges. Algorithms used to improve the images require more processing power, and better sensors increase power requirements or require more space on the interceptor. But processing power, electrical power, and space are often at a premium on tightly configured interceptors.

Numerica, however, has been able to make improvements in IR imagery because its algorithms can take a series of low-resolution images and convert them into one high-resolution image — and Numerica’s technology can do it more efficiently than competing tools, and without the use of a larger sensor system. Overall, the software is able to improve image resolution more efficiently than any other existing program, according to company officials.

How it Works

Numerica’s advantage lies in calculation techniques that do not take up valuable memory in a computer processor. This short-cut method executes in approximately 0.4 seconds on a 2-gigahertz processor. Thirty such calculations take 12 seconds of computation time. More common processing methods would require several hundred seconds for the same calculations, said Woody Leed, a software engineer with Numerica.

The company’s method involves acquiring a stack of low-resolution images, none of which are particularly helpful in guiding the interceptor to a target. Using the successive frames, Numerica’s technology can infer pixel data to generate a high-resolution image, said Dr. Randy Paffenroth, program director.

“Then I can take the high resolution [data] for all those things I want to use high-resolution imaging for — for targeting my interceptor, for deciding where it goes, for discriminating the target from debris, all these kinds of things,” he said.

To make the tracking even better, Numerica uses what is called multiple hypothesis tracking, which uses current information to correct past mistakes. For example, if an image was interpreted to be that of one object, but later after additional and perhaps more accurate images are acquired, analysis could determine that the image really shows two objects. “If I go back and correct all the things that I did thinking it was one object to now include the information that there are two objects there, it makes the accuracy of tracking these things much, much better,” Paffenroth said.

“Our algorithms allow you to tell that I was wrong before, but now I know what’s going on. So it lets me fix the mistakes I made previously, instead of just letting them propagate forever.” The net effect of this tool is to give the interceptor better information sooner so that it has more time to respond, and to improve the chance of a successful intercept.

The tool also could track pieces of missiles created when a successful intercept occurs, so that the uncertainty of where those objects will land is reduced, Paffenroth said.

While the company developed the software for use with interceptors, it also could be used with unmanned aerial vehicles (UAVs), unmanned ground vehicles, or small robotic devices that take images of objects but lack the processing power, space, or electrical power needed for high-resolution cameras. “If you just pine for high-resolution imagery, this is a way to finesse that,” Paffenroth said.

There are nonmilitary applications, too. It also could be used in other electronic devices, such as camera phones or magnetic resonance imaging (MRI) machines in hospitals. For example, MRI images of the heart are blurry because they need to be created over several seconds as the heart beats. With Numerica’s algorithm, low-resolution images could be resolved into a single high-resolution image of the heart, providing doctors with more clarity with which to diagnose a patient’s heart condition, Leed said. And for camera phones, images could be improved without costly hardware upgrades.

Although the software was designed and optimized for infrared data, using it for handheld cameras would require some modifications — such as retooling it to analyze differences in the camera’s multiframe sensor speed and accommodating color processing — but the core super-resolution algorithms would remain the same, Leed said.

Where it Stands

Numerica’s technology has yet to fly in an interceptor. Development work so far has focused on proving the technology in the lab, but the company is working to become a supplier to MDA’s efforts involving airborne infrared sensors, which seek to put IR sensors on UAVs so missile launches can be detected sooner than they are now detected by satellites, resulting in earlier intercepts, Paffenroth said. The company also is seeking a Phase III contract for this technology.

Following recent successful tests for their technology, company officials are continuing to work with prime contractors to develop a system that could be fielded.

More Information

For more information on the Numerica algorithms, visit http://info.hotims.com/34451-523  . (Source: Dale McGeehon/NTTC; MDA TechUpdate, Missile Defense Agency, National Technology Transfer Center Washington Operations).

---

---