Paul T.
Veteran
"Indeed, something called Nyquist's Theorem states that you would need to scan at at least twice the resolution to get the information in the original. Since Velvia 100F has an RMS grain of 9, we'd need something that can resolve at least half that, or 4.5 microns to get the information the slide has to offer. "
is that true? I offer this as a genuine question, because I genuinely don't know.
Nyquist's theorem, as I recall, refers to digital sampling of an analogue, ie continuosly varying, waveform. But is 'grain' continuously variable? Isn't it more like a 'clump' of information?
is that true? I offer this as a genuine question, because I genuinely don't know.
Nyquist's theorem, as I recall, refers to digital sampling of an analogue, ie continuosly varying, waveform. But is 'grain' continuously variable? Isn't it more like a 'clump' of information?
oscroft
Veteran
I think it works the same. I've just been thinking about an example...
Suppose you have a source image that has pixels that are black-white-black-white, and you sample it at the same resolution but your pixels are positioned off by exactly half a pixel. What you'll see is grey-grey-grey-grey (each sample covers an area that is half black, half white). But if you sample at twice the frequency, what you'll get is black-grey-white-grey-black-grey-white-grey. You'll have twice the number of pixels for the same source resolution (with the grey pixels in the sampled image being noise).
Exactly 50% offset is the worst case, but if you consider different offsets you'll get various combinations of light-dark-light-dark (which are various shades of grey). And in reality, film grains and pixels won't match one for one anyway, so you'll actually get all sorts of variations from black-white to grey-grey.
Does that make sense?
jlw
Rangefinder camera pedant
Sorry, but RMS granularity doesn't have anything to do with microns, which are a linear measure. It's a statistical measure (RMS stands for "root mean square") of density variation at a particular point on the film's sensitometric curve (it has to be pinned to a specific point because granularity varies along with density.)
If you don't mind using Wikipedia as a substitute for real research, you can read a more thorough explanation here (click.)
oscroft
Veteran
Interesting point. I suppose (non-expert speculation here) that to actually "get the information the slide has to offer" (as the article puts it), you'd need a scanning resolution equivalent to twice the finest grain resolution found in the image, not twice the RMS resolution.
oscroft
Veteran
And that using RMS grain size to compare film with digital makes film look worse (and digital look better) - the maximum resolution of film being determined by the smallest grain size found in the image, not the RMS grain size.
Have I got that right?
Have I got that right?
V
varjag
Guest
The sampling theorem has broad application in information/signal theory, not particularly specialised to digital, impulse or analogue signal. It is applicable when you have any form of sampling, i.e. encoding one signal with less precision/frequency than originally.Paul T. said:
Paul T.
Veteran
Yes, but my point is more about the information contained within those grains. You need to sample at double the frequency to retain that information, but that doesn't double the amount of information.
THe presence of a grain, or no grain, at one density, carries the same information as a bit, or no bit, at the same density.
THe presence of a grain, or no grain, at one density, carries the same information as a bit, or no bit, at the same density.
HAnkg
Well-known
Professional consumers of photography (ad agencies, cosmetic and fashion houses) who would not accept 35mm color film because it did not meet their quality standards are buying images from pro 135 DSLR's. I owned a color prepress and retouching service in the 80's when everything was still shot on film and then scanned. I had accounts like L'Oreal, Revlon, Cartier and Mercedes Benz. I can tell you from experience scanning 35mm film to 300MB is a waste of disc space. You might like the look of film and that's fine but as far as color image quality the professional marketplace that is most concerned about image quality has voted with their $$$. That discussion is long over.40oz said:
oscroft
Veteran
Indeed. And that's what makes it wrong to think that a film that requires, say, a 20Mp scan to resolve everything in it actually cointains 20Mp of information.
Yes, but you need two bits samples to detect one grain (or you might get the grey-grey-grey-grey result from my example). You are producing more apparent data (by having twice as many bits) but half of that data is noise and does not constitute information.
dazedgonebye
Veteran
Well, when Zeiss comes out with a digital RF, it'll be a complete piece of crap (read as "I can't afford it"). The M8 suffers from the same design flaw.
Bill,
I prefer cheese.
Bill,
I prefer cheese.
V
varjag
Guest
No, that doesn't double the amount of information of course - just the way sampling is. Same issue applies to digital camera sensors, otherwise you lose detail and get moire patterns.Paul T. said:
When we see though that a scanner resolves a line visually of 1-2 pixel wide it is however a pretty good indication that it's sampling rate (scanner resolution) that's the limit, not the film. We bump in the sampling frequency.
Gabriel M.A.
My Red Dot Glows For You
You had me at "to compete". What comes after just made me not believe the rest of the post.machineage111 said:
Gabriel M.A.
My Red Dot Glows For You
I hope so.ray_g said:
To tell you the truth...I'd prefer a full-frame Nikon dSLR over a full-frame Canon. Didn't see that one coming, did ya?
Huck Finn
Well-known
machineage111 said:
Not the way Zeiss does business, so the "rumor" doesn't hold water with me. Zeiss either licenses a lens design, which is then sold under the name of the manufacuturer (e.g. Rollei, Hasselblad, etc.) or they retain ownership of the product, control of the specs, quality control, warranty, marketing, etc. but subcontract manufacture to an OEM & sell it under their own name (e.g. Zeiss Ikon/Cosina, Contax G/Kyocera, etc.). They don't simply rent out or sell the rights for their name to be used on someone else's product or someone else's design. Completely different than what Cosina did in acquiring the rights to the Voigtlander name.