A 50 gigapixel camera five times better than 20/20 human vision

[Tim Gray]

http://www.kurzweilai.net/a-50-gigapixel-camera-five-times-better-than-2020-human-vision

The interesting part to me is the comment that pursuing sensor technology may be more productive in the long run that improving optics.

[BJL]

Looking at the blow-ups, sections of the image that are about 1/50 of thr total image are distinctly fuzzy (OOF? diffraction?) when displayed at less than 200 pixels wide on my screen, so the resolution is well under 10,000 pixels across, making the sample image well under 50MP worth, about 1000-fold short of the headline claim. Maybe the 50 GP refers to some proposed future upscaling: after all, to get 50GP from 98 tiny cameras would need each one to deliver about 500MP.

So I won't believe 50GP on the basis of anything in that article. The calculations of the giga-pixel camera project reveal some of the problems: atmospheric imperfections, diffraction/OOF effects and such make it very hard for any lens to deliver that much detail, except perhaps in an extreme wide-angle view of a subject entirely at a great distance form the camera, so that everyting is adequately in-focus with focus set near infinity even with a large aperture diameter. For normal field of view, controlling diffraction adequately would require an aperture diameter of about 100mm or more; doubling the angular field of view could get that down to about 50mm, and so on.

[tsjanik]

The reporters are not checking their facts.  50,000 MP from 98 tiny sensors implies over 500 MP/sensor. The actual number is 1000 MP @ 14 MP/sensor, in B&W only. The researchers plan to work on a 50,000 MP model (they best work on some new computers as well  ).

http://www.disp.duke.edu/projects/AWARE/index.ptml

[Ajoy Roy]

The concept is brilliant. It is as though you have 100 images shot and stitched simultaneously.

A similar principle is used in modern telescopes, where instead of one large monolithic mirror, multiple smaller mirrors are used to capture parts of the image and stitched to give one large image.

Provided the project culminates in commercial release, we can expect 4x5, 8x10 or larger size equivalent cameras. Imagine 100 1/2" x 1/2" sensors giving a total of 5" x 5" effective area or 300+ array resulting in an 8"x10" camera!

[Quentin]

Rubbish picture.  Whatever happened to aesthetics?

[ondebanks]

A similar principle is used in modern telescopes, where instead of one large monolithic mirror, multiple smaller mirrors are used to capture parts of the image and stitched to give one large image.

No - the multiple mirrors in a large telescope (like Keck or GTC) are all aligned to feed the same instrument. There is no image stitching arising from multiple-mirror segments. Put strips of black electrical tape across the front of your telephoto lens to divide it into "multiple lenses" and you'll see what I mean.

Ray

[ondebanks]

Looking at the blow-ups, sections of the image that are about 1/50 of thr total image are distinctly fuzzy (OOF? diffraction?) when displayed at less than 200 pixels wide on my screen, so the resolution is well under 10,000 pixels across, making the sample image well under 50MP worth, about 1000-fold short of the headline claim. Maybe the 50 GP refers to some proposed future upscaling: after all, to get 50GP from 98 tiny cameras would need each one to deliver about 500MP.

So I won't believe 50GP on the basis of anything in that article. The calculations of the giga-pixel camera project reveal some of the problems: atmospheric imperfections, diffraction/OOF effects and such make it very hard for any lens to deliver that much detail, except perhaps in an extreme wide-angle view of a subject entirely at a great distance form the camera, so that everyting is adequately in-focus with focus set near infinity even with a large aperture diameter. For normal field of view, controlling diffraction adequately would require an aperture diameter of about 100mm or more; doubling the angular field of view could get that down to about 50mm, and so on.

If you follow the links through to a detailed description of the instrument, it all hinges on one single - rather amazing, but still single - objective lens. I expect that's the limit to overall image quality.

Ray

[BJL]

... it all hinges on one single - rather amazing, but still single - objective lens. I expect that's the limit to overall image quality.

Agreed; that seems to be the great limitation on efforts to record extreme resolution in a single-time exposure rather than piece by piece sequentially. My skeptical thought is that the signal processing team has solved the problem of detecting all the fine spatial detail delivered by the lens, and partially dealing with issues of DOF and variation in subject brightness by adjusting the focus and exposure level "locally" with the separate sensors, but the project has a major challenge ahead in building a lens capable of delivering the advertised resolution. Maybe that is deliberately left to a later stage in the project, when a far bigger budget is available for the very careful lens fabrication.

It seems like another illustration of what I have been saying for years, that sensor and signal processing technology is quickly evolving to the point where the lenses become (once again) the central factor in the performance of a camera. Quite the opposite of the comment by David Brady (Professor of Electrical and Computer Engineering at Duke’s Pratt School of Engineering) that

The primary barrier to ubiquitous high-pixel imaging turns out to be lower power and more compact integrated circuits, not the optics.
--- http://www.kurzweilai.net/a-50-gigapixel-camera-five-times-better-than-2020-human-vision

which seems like a case of the "hammer-nail" syndrome.