Faculty & Research
Information Technology
Printable PDF

Research Profile

Open source camera gives researchers a platform to reinvent photography

August 2009. With a commercial digital camera you can only do what it lets you do. With no programmability or network access, it’s an instrument of creative expression whose internal workings are rigid and secret. That paradox strikes computer science Professor Marc Levoy as a shame, and so he and his students are fixing to blow digital photography wide open with a project they call “Camera 2.0.”

“The premise of the project,” Levoy says, “is to build a camera that is open source.”

Virtually every aspect of his custom-built camera, down to the lens, aperture, shutter, and flash, are exposed to the control of inspired programmers, giving them the freedom to experiment with new ways of tuning the camera’s response to light, focus, and motion. They can add their own algorithms to process the raw images in innovative ways. One well-understood example, is automatically stitching together images to create a larger scene such as a panorama. Right now, panoramic stitching is done offline in programs such as Adobe Photoshop. Levoy wonders why it couldn’t be done on the camera and why hobbyists couldn’t experiment with new ways to do the stitching.

Levoy’s plan is to develop and manufacture the device, dubbed “Frankencamera,” as a platform that will be available at cost to fellow computational photography researchers. In the young field of computational photography, which Levoy helped establish, researchers use optics benches, imaging chips, computers and software to develop techniques and algorithms to enhance and extend photography. This work, however, is bound to the lab. Frankencamera would give researchers the means to take their experiments into the studios, the landscapes, and the stadiums.

And perhaps someday to the birthday parties and on vacation. Extrapolating from the research world to the commercial market, graduate student Andrew Adams imagines future consumers downloading apps to their open-platform cameras the way people do with iPhones today. In fact, an effort parallel to the Frankencamera in Camera 2.0 is to bring the benefits of computational photography to cell phones, which may be lacking as cameras, but are blessed with what cameras lack: easy programmability and ample connectivity.

By either avenue, Levoy’s group stands to unleash a wave of academic and creative energy that could transform picture taking.

A panorama of ideas

For one example, among the most mature ideas in the field of computational photography is the idea of extending a camera’s “dynamic range,” or its ability to handle a wide range of lighting in the same frame. The process of high dynamic range imaging is to capture pictures of the same scene with different exposures and then to combine them into a composite image in which every pixel is optimally lit (see the images of a Zurich bridge below). Right now researchers do this with images on computers in the lab. Levoy wants cameras to do this right at the scene, on demand. Although the algorithms are very well understood, no commercial cameras do this today. But Frankencamera does.

Another algorithm that researchers have achieved in the lab, but no commercial camera allows, is enhancing the resolution of videos with high-resolution still photographs. While a camera is gathering low-resolution video at 30 frames a second, it could also periodically take a high-resolution still. The extra information in the still (especially that which doesn’t pertain to what’s moving in the frame) could then be recombined by an algorithm into each video frame. Levoy and his students plan to implement that on Frankencamera, too.

Yet another idea is to have the camera communicate with computers on a network, such as a photo-hosting service on the Web. Imagine, Levoy says, if the camera could analyze highly-rated pictures of a subject in an online gallery before snapping the shutter for another portrait of that subject. The camera could then offer advice (or just automatically decide) on the settings that will best replicate the same skin tone or shading. By communicating with the network, the camera could avoid taking a ghastly picture.

Of course users with Frankencameras would not be constrained by what is already known. They’d be free to discover and experiment with all kinds of other operations that might yield innovative results because they’d have total control.

“Some cameras have software development kits that let you hook up a camera with a USB cable and tell it to set the exposure to this, the shutter speed to that and take a picture, but that’s not what we’re talking about,” says Levoy. “What we’re talking about is, tell it what to do on the next microsecond in a metering algorithm or an autofocusing algorithm, or fire the flash, focus a little differently and then fire the flash again — things you can’t program a commercial camera to do.”

As many ideas as Levoy’s team may want to implement on the camera, the real goal is to enable the broader community of photography researchers and enthusiasts to contribute ideas they couldn’t possibly imagine. The success of Camera 2.0 will be measured by how many new capabilities the community can add to collective understanding of what’s possible in photography.

Behind the lens cap

To create an open source camera, Levoy and the group cobbled together a number of different parts: the motherboard, per se, is a Texas Instruments “system on a chip” running Linux with image and general processors and a small LCD screen. The imaging chip is taken from a Nokia N95 cell phone, and the lenses are off-the-shelf Canon lenses but they are combined with actuators to give the camera its fine-grained software control. The body is custom made at Stanford. The project has benefited from the support of Nokia, Adobe Systems, Kodak, and Hewlett-Packard. HP recently gave graduate student David Jacobs a three-year fellowship to support his work on the project. Kodak, meanwhile, supports student Eddy Talvala.

Within about a year, after the camera is developed to his satisfaction, Levoy hopes to have to have the funding and the arrangements in place for an outside manufacturer to produce them in quantity for ideally less than $1,000. Levoy would then provide them at cost to colleagues and their students at other universities and, of course, have a number available for research and teaching at Stanford.

The son, grandson, and great-grandson of opticians, Levoy sees his mission as not only advancing research in computational photography, but also imbuing new students with enthusiasm for technology. This spring he launched a course in digital photography in which he integrated the science of optics and algorithms and the history of photography’s social significance with lessons in photographic technique. The class was a hit, with 160 students enrolled [see their best photos].

Who knows what any one of those students in a few years might enable a truly open source camera to do?