Knowing the angles that different rays of light travel allows the camera’s software to simulate the photo that would be produced by a virtual cameras focused in a particular way. When a person interacts with a Lytro photo, software tweaks the settings of that virtual camera to produce the new, refocused image.
Lytro’s sensor is made by bonding a carefully etched sheet of glass on top of a conventional digital-camera sensor. The glass is patterned with tiny lenses, ensuring that specific pixels can receive light only from the specified angles. That gives Lytro’s software the information it needs to refocus photos.
Another consequence of this design is that the camera records depth, which makes it possible to reproduce 3-D images. “We’re not going to be emphasizing it from the start, but these pictures are inherently 3-D,” says Ng, who showed Technology Review images from a Lytro camera on a laptop with 3-D-capable screen.
Lytro’s approach to camera design and photography emerges from a relatively young area of research known as computational photography. Researchers in that field use various computing and mathematical techniques to achieve novel feats of photography and videography, including taking cell-phone photos in very low light or even taking pictures around corners.
Ramesh Raskar, who heads the computational photography research group at MIT’s Media Lab, says that Lytro is the first company to try to commercialize computational photography. “The camera industry looks at what we do as very new and experimental,” he says. “If Lytro are even partially successful, they will make people realize that computational photography can be practical.” Raskar says that Lytro’s basic design approach is sound and that he believes users of conventional cameras will be interested in the ability to focus after the fact.
However, Raskar adds that Lytro’s sensor design causes its output to be of lower resolution than an equivalent sensor configured normally, because of the need to restrict pixels to receive light only from certain angles. Raskar’s own research group have an alternative design that places a sheet perforated with small holes slightly in front of a camera’s sensor. That arrangement doesn’t have the effect of specializing pixels to certain directions of light as in Lytro’s sensor, but it does attenuate light rays in a known way such that the path of different light rays can be mathematically worked out from what the sensor records. An image can then be refocussed as with Lytro’s design.
Most importantly, the MIT lab’s approach cuts the resolution of photos less, and in a way proportional to the amount of depth range a person chooses to be in focus, says Raskar. By contrast, Lytro’s resolution penalty is always the same and likely means a cut of at least ten times in a sensor’s output in each dimension, he says. Raskar says there is strong interest in commercializing his group’s design, although he is far from ready to launch a competing product to Lytro’s.
Ng says camera sensors are today so high-resolution that any resolution penatly should not be a problem. He argues that marketing efforts by camera manufacturers have led to consumers to believe they need more megapixels than they do. “Most of photos that are shared are a tiny fraction of a camera’s ability,” he says. Ng wouldn’t say what the output quality of Lytro images is, preferring to say that his sensor captures 11 million light rays of data (or 11 “megarays”). The largest images shown by the company online are 800 pixels square. A standard six-by-four-inch photograph requires a digital photo that is 1,800 by 1,200 pixels in size.