Plenoptic Cameras

[Tim Gray]

An interview with these reasearchers might be an interesting VJ segment.

http://graphics.stanford.edu/papers/lfcamera/

[John Camp]

That's great, but does it go to 3200? I absolutely couldn't use it if it doesn't have noise-free 3200.

JC

[Tim Gray]

My layman's interpretation is that noise would be inversely proportional to the ratio of the resolution of the sensor to the resolution of the final image.  So if the sensor was a 40mpx and the microlenses ratcheted that down to a final image of 10 mpx then you'd have 4 times less noise at 3200 than with a conventional camera.

In this paradigm you want more photosites - and the smallness of the sites doesn't drive the noise, the noise is a function of the size of the microlenses (subject to the diffraction limits of the aperture).  

[BJL]

I think I agree with Tim: this could combine thw low noise advantags of large apertures with the greater DOF of small ones. Noise levels are ultimately limited by how much light you gather from the subject: more signal, better signal-to-noise ratios. So when this camera gathers lots of light by using a largish aperture and then rearranges the output of the photosites to get more DOF or a different focal plane, it keeps that good SNR.

[lester_wareham]

An interview with these reasearchers might be an interesting VJ segment.

http://graphics.stanford.edu/papers/lfcamera/
[a href=\"index.php?act=findpost&pid=51628\"][{POST_SNAPBACK}][/a]

Yes I saw this, the maths was quite interesting. Considering the 4D fourier transforms needed I don't think we will be seeing in camera processing for while but if enough Mp sensors become available it could be good for post processing.

I didn't work out how long the processing would be on a bog standard 4GHz PC, did you?

[Tim Gray]

Maybe you could set up a distributed computing network like they did with Seti

It's interesting, since the tendency is to think -  how could I possibly require more horsepower? - and then something like this comes along.

[crspe]

I saw this article about a week back.  I am quite sure that this will never see a practical application ... They take a 16MP input image to produce a 0.08MP output image!!! That is a scary loos of reolution.  The whole concept only works with a fixed aperture - they use an f4 lens with matched microlenses ... if you go wider than f4 then the subpixels merge together and you cannot get an image out of it.  if you stop down then you lose the whole benefit of the microlenses ...

Note that they also are only 50% efficient in gaining DOF with this concept ... they get a factor 7 improvement in DOF for a factor (14 * 14) reduction in resolution.  

A cheaper, easier, better quality way to get an improvement in DOF is to use a sensor that is a factor 7 smaller (and use a factor 7 wider focal length). Any 1/1.8" sensor would do.  The big advantage is that this is a 100% efficient way of improving DOF, so you can choose to stay with a 0.08MP image but increase the pixel size by a factor 4 (in area) so get better light response, or stick with the same size pixel and get a factor 4 increase in resolution - 0.32MP, VGA!
Better quality, factor 50 less cost, factor 10 less weight.

Interesting article, no practical application.

[rogatmarois]

 I was excited to see this article.
http://www.newscientist.com/channel/info-tech/mg18825255.000
The research is in its early stages and many limitations are evident. It would be great to read an interview with these researchers.

Some questions for the researchers would be:

Do you see an immediate application for this?
What are the limiting factors for this "focus pulling" technique?

I see a possible application in security cameras. Imagine focussing on the suspect after the picture is taken.
Could we do away with auto focus or focus altogether in cameras and choose our focus subject or DOF in the digital darkroom? Would it be possible to make the whole photo in sharp focus?
Would it be possible to use in a video camera?

[BJL]

I saw this article about a week back.  I am quite sure that this will never see a practical application ... They take a 16MP input image to produce a 0.08MP output image!!! That is a scary loos of reolution.
[a href=\"index.php?act=findpost&pid=51761\"][{POST_SNAPBACK}][/a]

Probably no application to normal photogaphy, but it has potential for security cameras, since it can produce enough resolution for that purpose with a combination of good low light performance and effectively great DOF. The processing to effectively change the focus distance can also be used to get an image that is sharp over a wide rage of subject distances.

The need to used a fixed largish aperture is not really a problem since the post-processing overcomes the shallow DOF that is the main disadvantage of large apertures.

[crspe]

Even for security applications it has no application - instead of using this technique to get a factor X improvement in DOF, use a sensor that is a factor X smaller (and lens that is factor X wider).  With identical aperture and pixel size (ie same low light performance), you will get the same (improved) DOF, a factor 4 improvement in resolution, all for 1/10th the cost or less!!!

[BJL]

instead of using this technique to get a factor X improvement in DOF, use a sensor that is a factor X smaller (and lens that is factor X wider).[a href=\"index.php?act=findpost&pid=51848\"][{POST_SNAPBACK}][/a]

You might be right: to get that factor X improvement in DOF, you would need to shrink focal length and sensor size by factor X while keeping the same aperture ratio. Then the same shutter speed would require the same ISO and so to maintain noise levels would require the same photosite size, which sounds like it gives the same resolution in lines per picture height, unless I am misunderstanding the description of the resolution of the reconstructions.
If so, the only advantage might be getting the various different shallow DOF options. Fans of big, blurry balls of background bokeh could blend a massively OOF background image from reconstructing at a short focal distance with an in-focus reconstruction of the main subject.