Insects with compound eyes have a different perceptual apparatus than bifocal vision creatures like humans do, and their manner of perceiving the world is very different from our own. Looking at a raw light field camera "negative" is a bit like looking at a representation of a fly's vision: a mess of tiny, seemingly indistinct spherical renditions of a common scene. To assemble this raw data into a result deserving of the name "photograph" requires a unique process. Think of it as the plenoptic equivalent of developing film.
Lytro's software is based on Ng's doctoral research in viable methods for plenoptic image capture. The device combines each of the microlens perspectives on the scene into a single result. Because each microlens only records visual data on a small segment of the digital sensor, a Lytro image is much lower in resolution than a typical digital photograph. But in exchange, the Lytro produces not one image, but a set of virtual cameras capable of rendering a whole field of possible images, which can be configured and combined via software.
So what, though? What can you do with a light field photograph that you can't do with an ordinary one?
In the computer science research community, plenoptic photography is often suggested as a darkroom technique meant to assist the photographer in the production of a final, traditional, static image. And such a use case is very much possible with Lytro's software, even if the resulting static image is mostly good for screen display or small prints due to its low resolution (1080x1080 pixels) compared to today's consumer digital cameras. But Lytro's intended use is surprising: instead of using the light field data to "develop" a final image, they offer a Flash-based web viewer which allows a user to actively interact with the image, manipulating it live.
But how does one interact with a light field photo? For one part, such images are refocusable in post-processing. Go ahead and try it with the image above, changing the focus to the different tealight candles on the table. In his dissertation, Ng discusses a common case in which such a scenario would be desirable. When taking a portrait, it's common to use a large aperture to produce a shallow depth of field to isolate the subject. But given such a narrow margin of error, it's easy to misfocus due to subject or photographer movement. In this case, the decisive moment might have been captured -- a particular facial expression -- but focused too far: at the ear instead of the eye, for example. A light field photographer could correct this fault and produce the desired focal plane through a virtual camera in the Lytro digital darkroom.
This sort of example has given Lytro's the reputation of a "focus-free" camera, but that's not really an accurate depiction. After all, Cartier-Bresson's Leica images were largely focus-free, since he zone-focused at smaller apertures to insure that most of the scene would be in focus anyway, allowing the photographer to concentrate on anticipating and capturing the decisive moment. Likewise, point-and-shoot cameras and many cell phone cameras have small focal ratios thanks to very wide lenses, making them shoot almost an entire scene in focus anyway. Focus-free is different from refocusable.
The result offers a kind of visual pun or reveal that forms the current Lytro aesthetic, for better or for worse.
Even given that caveat, Lytro light field images aren't fully refocusable -- that is to say, all the possible virtual camera configurations in a living negative don't correspond with all the possible real focus configurations of all possible traditional cameras. This is because the light field the Lytro captures is not in the world, but in the camera's dark chamber.
That means that the light field itself is partly a result of the camera's primary lens focal point. Even though the lens can open up to f/2 up to 300mm at its longest zoom, closer focusing still produces a shallower depth of field. The result is that Lytro photos software-focused to objects in the foreground produce shallow depth of field all throughout the image, while those with objects further in the distance produce less depth of field. For this reason, many of the more effectively "refocusable" Lytro images deploy a kind of one-two punch delivered by an object in focus in the very near foreground, which occludes some other object in the background. Clicking on this far object brings it into focus as a kind of visual punch-line.
For example, this image of a Scrabble board is addressed to someone in the foreground. The tiles in the background bokeh are clear enough for the viewer to discern them, and clicking on them shifts focus to reveal a message. The result offers a kind of visual pun or reveal that forms the current Lytro aesthetic, for better or for worse.
Images focused at a medium distance tend to deplete focal distinction due to the increased natural depth of field at further focused distances. This image of a girl playing with a bubble toy illustrates the effect:
The image is set to load at a very close focal distance, that of the bubble gun (A Lytro "living image" can be configured to choose any virtual camera as a starting point of focus). But clicking on the bubbles in the middle of the frame displays microlens data focused on a farther distance from the "natural" point of focus, an image whose native depth of field is simply greater. The resulting action just isn't very interesting. Cinematic techniques like focus in, focus out transitions have set conventions for disclosing a subject over time. But in Lytro's compromise with this filmic technique, the promise of revelation implied by the primary, out of focus portion of the image falls flat. It turns out to be just a mediocre snapshot.