plenoptic cameras

[emanresu]

Just saw this interesting thing.  A prof from our school is giving a talk next Friday on this topic.  The most exciting part is that the cameras are already on the market!  Maybe its time to stop buying regular DSLRs and save money for these cool cameras.

http://www.tgeorgiev.net/

[emanresu]

I attended the talk and it was very interesting.  They customize-made a grid of microlenses, each is square in shape (for easy alignment) and projects to a 75x75 pixel area on the sensor.  They used a phaseone digitback and replaced the IR filter with the microlense and took picture.  the unprocessed image looks like a picture seen behind some privacy glass tiles, but that RAW has to be processed.  The professor said that they originally used HPC clusters to do the processing with CPUs, and it would take 5 minutes to render just one frame (only 3-4 MPs).  But now with GPU and CUDA, they are able to render 12 frames  per second!!   He said the code for processing is not hard at all, mostly using the shader function, and even did a live demonstration!   So it is definitely a practical thing, the only thing that matters is the cost.

[Bob Atkins]

Isn't resolution reduced? Aren't you trading off resolution for the ability to "focus" after the fact? If you have  50 micro-lense and array and each element uses 1/50th of the sensor area, don't you end up with an image that's 1/50th the size of the whole sensor once you've done all the processing? Maybe it's a little bigger than 1/50th because the individual images will image slightly different areas due to their different views, but even allowing for that I assume the final image has to be a lot smaller than the sensor size.

I haven't looked at it in detail, but if you look at it in terms of imformation theory, I'd assume you have to trade off information about spatial resolution for information about object position (focus).

I guess if you start off with an 80MP phase 1 digital back you can still end up with a reasonable size and resolution image.

I can see applications for the technique, but I'm not sure that photography for art or documentary work is one of them. Maybe for surveilance video or scientific imaging?

[emanresu]

Hi Bob,

You've hit it spot-on.  It does sacrifice resolution for information, and that is why the final image is only 3-4MP despite they used a phaseone digiback.

I personally think it is an interesting technology, but quite frankly, after attending the talk, it didn't seem quite as magical as it sounded on the poster.

The poster advertised that it could achieve HDR, and could achieve different focal planes, different DOF, but HDR is just a theory because it still uses the same image sensor as any other digital camera, and the theory is that they can add a grid of ND filters (with different grades) in front of the micro-lens array, but I guess we can do that with HDR much more economical and easily.  And the focal planes are just simulated, and so is the DOF.  I guess that because the microlens are so small, they are effectively lenses with tiny apertures, and give a very deep DOF, and the software just later blurs things out when shallow DOF is desired.  In fact, I can hear Ken R. screaming because this will further promote postprocessing using computers instead of photography using cameras.