[mozumder]

Which is why I said you need to distribute the rays around randomly in a gaussian distribution. Don't just arrange them in perfect rows and columns - that's the worst array option that guarantees aliasing. You can shift each pixel (or subpixel) lens around slightly in a permanent pattern.

Anti-aliasing techniques are common elsewhere in 3-D graphics, and can be used just as well in light-fields.

[ryandamm]

That assumes your pixel distribution has much higher potential addressable resolution than the signal you care about... in which case, why not just display the full raster?

Rendering for 2D and display are two different beasts -- but I'll own the fact that I'm not formally trained in this and there may be subtleties -- or even obvious signal processing facts -- that I'm getting wrong. (But if I'm wrong, I'd love to know how, for my own edification.)

[mozumder]

> That assumes your pixel distribution has much higher potential addressable resolution than the signal you care about.

It doesn't need that at all. Light fields can be interpolated like anything else, just like bayer filter for color on camera sensors or 4:2:2 color compression on the signal side. But if you're doing 3-D rendering, you can match rays exactly to your distribution on the display if the renderer knows the distribution of rays on the display.

Interpolation is always going to reduce quality, but it's better than aliasing, so there's going to be a trade-off analysis that needs to be done. I don't know what the results of that would be, so this is all theoretical.

[ryandamm]

With all due respect, interpolating light fields is far less trivial than you make it out to be. It's a 4D field, and naive interpolation leads to loss of detail, and often edge doubling (itself a form of aliasing).

Furthermore, if you're interpolating rays, you're necessarily not doing what you originally proposed, which is to only light up a (random or pseudorandom or evenly distributed) subset of the pixel display elements, presumably to save on rendering cycles.

Let me just say, more generally, that intuition trained on 2D doesn't apply directly to light fields.

[mozumder]

Tru dat. Interpolation will probably be too difficult in 4-D space.