That said I suspect that eventually a stochastic process could give the same result as a deterministic process if that's what you mean.
Scanning a poor photograph can't make it a good photograph no matter how high the scan quality is.
FWIW...The result of a stochastic process in all but trivial cases is a random variable(s). Thus, it is not deterministic. However the parameters of the resulting random variable(s) can be deterministic. For example, the expected value of the stochastic process can be deterministic (but it can also be a random variable itself).
A comment on the original topic... Based on Tim Vitale's article (which is worth reading) the approximate digital resolution of 4x5 Tmax 100 would be 9,700x12,200 pixels if your lens could resolve 70lpm. That's 118mp. This would equate to a 30"x40" print at 300dpi. Lenny, how do you get 320mp off of 4x5?
I would like to see a valid large print (30x40+) comparison as well; maybe 5DII, Phase One P65, and Top quality Drum Scanned 4x5 Tmax 100 and Velvia.
You can get (roughly) a 320 mp file from 4X5 by scanning at 4000 spi.
How much of that is useful pixels is debatable. My experience is that there is no gain from scanning at more than the effective resolution of the camera/lens/film system. Scanning at 4000 spi suggests that the scanned material has the equivalent of about 80 lines per millimeter of information. 80 llines per mm on 4X5 may possible in certain specific conditions, but in most practical situations one is very lucky to get as much as 50 lines per mm on film, which is about 2400 spi. That gives a file of about 115 mp, which IMO is about as many effective mp one can get from 4X5 format.
Yes, this is a very misunderstood topic. Scanning for more resolution than the camera/film system can deliver is proving dubious with experience. Beyond that it gets back to grain representation. Then how is that definition delivered? In the case of a drum, despite common thinking, it's the aperture not file size. Up to a point of course.
Then, to get to a big print, how one gets to the printer native ppi is the real question. More physical sampling from the scanner, higher than that native for the chosen aperture? Rezing in editing software? Letting the print driver or RIP do it?
Lot's of interesting stuff here-
I've drawn no conclusions from that yet, but a few tests show the presentation is viable.
I think useful pixels is certainly debatable. However, my experience doesn't match your conclusions. I think the difference may be in the scanner, or more specifically, PMT's vs CCD's. However, if you are correct, then an Epson 750 at 2400 would yield the same file as a top drum scanner (or even a Cezanne). I just don't see this to be the case. Every time I look at a scan from these animals, they don't match up.
Maybe I am wrong. Do you have a test negative that you use to determine this? I would be happy to scan it here at different resolutions and put it to the test.
That said, I am reminded that I am always frustrated by these tests. Resolution is one thing. Sensitivity, pixel by pixel, is quite another.
And just in case you are wondering I'm not trying to piss you off - I sincerely would love to know the answer...
Museum Quality Drum Scanning and Printing
"However, if you are correct, then an Epson 750 at 2400 would yield the same file as a top drum scanner (or even a Cezanne). I just don't see this to be the case. Every time I look at a scan from these animals, they don't match up."
Absolutely not. I am not saying that at all, and indeed am surprised at your confusion. I think we agree that some scanners deliver more real, effective resolution than others. Drum scanners, if calibrated, should deliver close to 100% of stated optical resolution. My EverSmart Pro delivers close to 95% of stated optical resolution. Most consumer flatbed scanners, Epson 4990, Epson V750, etc. deliver only about 40% of stated optical resolution. You can test this by scanning a high resolution target: you may have a pixel count of 4800 spi, but the effective resolution would be only about 40% of this, say about 40 lp/mm.
It is not a question of PMT versus CCD technology in the least. It is simply a question of understanding the fact that all pixels are not equal in terms of their detail/resolution. The concept of a "useful" pixel is only debatble if one has no understanding of what "useful" is.
For example, if you want to make that 50 x 40 inch print from 5x4 film. You could scan at full file size. Or, you could scan at effective resolution which would give you say a 40 x 32 inch print, and then uprez to 50 x 40 inches.
The little bit to be gained here is the elimination of the uprez step. Because every step that requires some shuffling of pixels to some extent looses some information.
Now is that little bit of gain visible in the final print? YMMV. On the other hand, once you've got the film on the scanner, why not? Again, YMMV.
Anyone care to comment on how the a sampling frequency of between 2600 and 4000 ppi can be the wrong match for the grain structure of specific films? I've run up against grain aliasing/moire with color print films like NC and Portra which seems to amplify the appearance of their grain structure. Wouldn't this make a strong case for sampling at higher ppi--even if there were no gain in line pair or edge resolution to be found?