Some of you are having trouble figuring out exactly what is and isn't possible with Light Field processing as proposed by Lytro. I thought Ng's dissertation was reasonably clear, but let me step through one simple example to give you an idea of what you gain and what you give up.
Let's take the typical 10mp BSI compact camera sensor at the moment. In front of that we're going to put an 4 by 4 array of microlenses to do the Light Field type collection. We'll use an f/2.8 lens with a fixed focal length of 8mm. This gives us the equivalent of an f/11 depth of field (you multiply the aperture by the square root of the array pixels, of which we have 16). By prefocusing the lens at 3 feet we get 1'5" to infinity in focus. That's theory. In practice Ng claims to be getting half that, but still, it's clear that we should be able to get a wide focusable range with a fairly small array. The downside? We've got images that are about 700 kilopixels. Decrease the array to 3 by 3 and we have an f/8 lens focus equivalence and 1.1mp camera.
Some people have predicted that the megapixel race will end, that we're out of gain for the effort. Light Field technology says no, it will not. We get noise reduction benefits from collecting so many samples for a single pixel, plus we get focus benefits, as well.
The question being raised is whether this is a disruptive technology or not. Like the Foveon sensor, which effectively increased pixel count per photosite, there appears to be an immediate potential gain over traditional Bayer if you scale it big enough. The question is whether Bayer can continue to brute force its way to market dominance. At the high end of camera range, probably. But Light Field is much more interesting at the cell phone camera size. It's not difficult to imagine an 8mp cell phone sensor with Lytro's array producing very high quality, no focus necessary images for Web use (Facebook, Twitter, Blogs, Flickr, email, etc., which is where most camera phone photos are going).
As one reader wrote in an email to me (thanks Martin), the most likely disruption is to the lens makers. This is certainly true at the cell phone end, where we've got people trying to cram zoom and focus systems into very small spaces. But Light Field has potential to disrupt a lot more than that. One of Nikon's primary claims about Coolpix cameras, for instance, is that it is their lens that makes the difference (a huge number of compact cameras share sensor and sometimes even more internal design). What happens when you don't need as complex a lens and have additional data (the Light Field array information) that can be used to correct lens problems?