Smartphones have fundamentally changed the world: Information is more readily available, computer sales are declining, and most people now have a high-quality camera in their pocket.
As pocket computers become ubiquitous, the demand for high-quality components that can fit into a handheld device has skyrocketed. The availability of cheaper, more powerful camera sensors and mobile processors is what made it possible for four-year-old startup Light to build its $1,699 L16 camera using not one, but 10 different lenses—kind of like how a spider has eight eyes. Light CEO Dave Grannan says the L16 is on par with a professional, full-frame camera (which is about twice that price) but Quartz hasn’t had a chance to test it.
Grannan says the L16’s quality is possible because the camera is built around computational photography, a burgeoning computer-science field that algorithmically constructs photos from data, rather than capturing exactly how light falls on an image. We spoke to Grannan about the L16 and how the concept of the camera is changing.
How does computational photography work?
Computational photography is any time you’re using multiple exposures and altering them with software. There are two inflection points that are allowing computational photography to go mainstream. Those are the ubiquity of very small, inexpensive cameras and processors that were really brought to us by the development of cell phones over the last six or seven years.
A big camera has a big sensor, where each pixel collects a lot of light. But it turns out that the laws of physics makes that additive: If you take 10 pictures and 10 sensors-worth of light, adding those 10 sensors up is exactly like having a great big sensor that would require a big lens. Sensors on small cameras don’t have a lot of light-gathering capability, and you need a lot of light for a great picture.
We get around that by using multiple small image-sensor cameras. We’re just reaching the point where mobile processors are able to run the computational image-processing algorithms, to take multiple exposures—in our case we take 10 pictures at once—and combine those computationally into one high-quality photo. So we’re right at this point where you have small, inexpensive sensors and computational power have come together to enable this for a consumer product.
What’s the promise of computational photography?
Imaging is the future of communication. Everything we do is visual communication, and computational is the future of that.
Everybody likes those artistic-looking photos, even if you’re not a photographer. We create a shallow depth of field, not traditionally by a great big lens that has an adjustable aperture, but because we take 10 photos, we can shift those photos in the background to be lined up with those in the foreground, and shift the background images so we get that nice background blur without requiring a great big lens.
I think the other thing that’s really exciting is, [Light] is a 3D camera. One of the things that we intend to expose is that we’re opening our file format, the Light Raw Image, the LRI file, to developers. So if you take a picture of a room you want to decorate, a developer could take our file format and write the application that takes the photo and draws a map of the room with measurements as precise as +/- one millimeter.
Think of e-commerce: Say I need a new appliance for my countertop and I don’t know what the measurements are. I just upload my photo to Amazon and say, ‘show me microwaves, but only ones that fit in the space under the countertop.’
We’re actually working with a telemedicine startup that wants to use cameras; the problem with telemedicine is that it’s very hard to determine the size of something from a photo. People try to use reference things like lay a pencil or a coin next to whatever they’re taking a picture of. But even then it can be hard to tell if there’s swelling or if there’s a wound getting bigger or smaller. We can actually measure that with our camera.
How do the 10 cameras measure distance?
It’s what we call parallax. If we know how far apart our cameras are, we can triangulate where the images line up pixel by pixel, and then try to overlay them. It’s just a triangulation calculation. We see depth with our human eyes because we have two of them. Two cameras can give you fairly good depth—10 can give very precise depth.
This interview has been edited and condensed.