A new imaging technique turns motion blur into an advantage, using a jiggling camera and a clever algorithm to create super-resolution images sharper than would be possible with a steady camera.
PROVIDENCE, R.I. [Brown University] - It doesn't take an expert photographer to know that the steadier the camera, the sharper the shot. But that conventional wisdom isn't always true, according to new research led by Brown University engineers.
The researchers showed that, with the help of a clever algorithm, a camera in motion can produce higher-resolution images than a camera held completely still. The new image processing technique could enable gigapixel-quality images from run-of-the-mill camera hardware, as well as sharper imaging for scientific or archival photography.
"We all know that when you shake a camera, you get a blurry picture," said Pedro Felzenszwalb, a professor of engineering and computer science at Brown. "But what we show is that an image captured by a moving camera actually contains additional information that we can use to increase image resolution."
The study was presented recently at the International Conference on Computational Photography and is posted on arXiv.
Digital cameras produce images by averaging the intensity of light over an array of pixels - tiny squares arranged in a grid. This sets a resolution limit: details smaller than a single pixel get smeared out across the pixel rather than precisely located within it. That causes sub-pixel details to be blurred.
The technique developed by Felzenszwalb and his team uses camera motion to produce sub-pixel resolution. When the camera moves, small points of light leave tracks that cross multiple pixels. The team's algorithm uses those tracks as extra information to pinpoint exactly where fine details must have been, reconstructing them on a finer grid. The result is a super-resolution image with detail sharper than the original pixel array allows.
For the study, the researchers tested the technique by mounting a conventional camera to a moving stage, which enabled them to test their techniques in various movement scenarios. In some cases, the team took multiple photos while moving the camera slightly between exposures, then used their algorithm to construct a single image from the multiple shots captured by the camera between movements. In other cases, the team moved the camera during each exposure and reconstructed a higher-resolution image from a single motion-blurred shot.
In both cases, the team showed that their algorithm could harness the camera motion to produce images with far higher resolution than would be possible without the motion.
"There was some prior theoretical work that suggested this shouldn't be possible," Felzenszwalb said. "But we show that there were a few assumptions in those earlier theories that turned out not to be true. And so this is a proof of concept that we really can recover more information by using motion."
The researchers envision plenty of potential applications for their technique. A moving stage setup like the one used for the experiments could be used for super-resolution archival photography of artworks or artifacts, the researchers say. The technique could also be useful for photography from moving aircraft.
The team also sees a possibility for the algorithm to one day run on commercially available cameras.
"There are existing systems that cameras use to take motion blur out of photos," Felzenszwalb said. "But no one has tried to use that to actually increase resolution. We show that's something you could definitely do."
The team plans to continue developing their technique and look for industry partners to make it available to the public in the coming years.
