Sorry, I just don't think Canon can do it

Started Feb 21, 2012 | Discussions thread
jayrandomer
Contributing MemberPosts: 600
Like?
Re: How much oversampling is enough?
In reply to John Sheehy, Feb 28, 2012

John Sheehy wrote:

jayrandomer wrote:

I guess the real question I have is, "how much oversampling is enough?" Three times? Eight times? Four thousand times?

What is a "time"?

It takes 4 pixels to do a half-@ssed line pair, 6 to do it with very low distortion/aliasing. That's for luminance. For a distortion/aliasing free blue or red channel with Bayer (red or blue-filtered B&W image) you're going to need 8/12. There is very little to be gained beyond this.

Either you don't think there's a diffraction limit (which was the claim I was clumsily trying to rebuke)

I don't like to use terms that are easily misunderstood. What diffraction does, is it causes an angular displacement of the simplified ideal trajectory of a photon/wave. The smaller the aperture, the more the displacement. Large pixels have a very great power for such displacement, and pixel size always displaces additionally to fraction alone, or diffraction and motion blur combined.

For someone who's so picky about referring to a multiple as "time" that's a clumsy description of diffraction. Diffraction is a host of effects related to the wavelike propagation of light. Spreading of beams through an aperture is one of those effects, but so is constructive interference from periodic structures (like the grooves of a CD) and the Poisson's spot at the center of a shadow.

or at some point higher pixel density isn't worthwhile. I agree that even with a diffraction limit, to lowest order an increase in pixel density will always bring a finite increase in image resolution but that increase becomes increasingly smaller. At some point you are just resolving the blur better (in microscopy called empty magnification).

If you don't believe me, would you at least believe Nikon? http://www.microscopyu.com/articles/formulas/formulasmagrange.html .

I don't have time to read all the other pages, but the 'a' and 'b' examples at the top are bogus, because there is no equality; the allegedly over-magnified image has the same subject size as the crisp one; how in the world is that possible in the question of practical magnification? The comparison is a fraud. If it was a real comparison, they'd show the crisp one much smaller, then upsample it to the same subject size as the allegedly over-magnified version.

I can understand not having to read all those pages, but if you had read further down on that page you would have seen why that image wasn't necessarily a fraud. A high magnification, high numerical aperture objective with a low magnification eyepiece will give the same magnification as a lower magnification, lower numerical aperture objective with a higher magnification eyepiece but much greater detail. You probably don't care that much about microscopy, though.

I believe what I can test myself. This world is full of quacks, or people who do not know how to communicate effectively. Even people who are correct are misquoted out of context. Look, we have cameras with pixels as small as 1.38 microns available with RAW, and they do not oversample at f/8. I get aliasing in the blue channel of my 1.5u camera at f/7.9. Diffraction is real, but diffraction is also a boogie-man; there are far greater ways to lose detail than diffraction, large pixels being the primary cause.

For digital photography diffraction may be a boogieman; I agree we've got a ways to go before resolution becomes wasted--a factor of 8 in size is a factor of 64 in MP. Diffraction is very real and at some point, hopefully in my lifetime, we'll reach a limit where it makes no sense to increase pixel densities further.

Did you do my simulation?

I did a simulation, blurring and pixelating until I found a pixel size that was sufficiently similar to the next size up for a given level of blur. You can certainly tell the difference at all levels of pixelation, but at some point you're not noticing any real improvement, only a difference. My conclusion from that exercise was similar to your conclusion above.

Single sampling is most certainly not that point and I profusely apologize for claiming such in haste (mostly out of habit from doing everything to an order-of-magnitude), but presumably there is some limit of oversampling that isn't worth even the added storage costs (and certainly not the added sensor development costs). I agree we're not there yet and won't be in the foreseeable future, but that limit exists.

What in the world is "single sampling"? If you're talking about sampling at the nyquist, that is impossible to do properly.

No, just talking about sampling a pixel size that's equal the diffraction-limited spot size. It's the terminology that Nikon used in their article on digital imaging sensors and microscopy.

-- hide signature --

John

Powers of 2 pixelation horizontally and blur vertically.

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow