I was unable to post these thoughts in the 'crop factor' thread before it degraded so here goes:
I have been investigating what would be the best prime for DX and came to the following conclusion: no glass is sufficient to resolve the sensor's (24Mp) detail.
Think about it; 4000 pixels over 15.6mm. That's 256 lines/mm, or 6500 lines/inch! Back in my old film days we were talking about perceptive limits of 100 lines/mm and now we are over double that!! What glass is capable of resolving to that?
Now on FX sensors, it is closer to 4000 lines/inch, so all things being equal, you will perceive more pixels... but still, it seems the glass just can't keep up.
So is there any point to more pixels on DX (or even a FX) sensor plane? I imagine there is still many areas in which sensors can be improved upon, but is more pixels one of them?
Do you think we have hit the wall?
I don't think we have hit a wall. I'm not sure there is any wall. But as long as we keep saying Wow about the new sensors, we're not at a wall yet.
I assume your "perceptive limits" meant 100 line pair per mm. Resolution of film and lenses generally are in line pairs.
Black lines have white lines between them, which is a pair of two lines, both of which have to be resolved.
See this:
http://cool.conservation-us.org/coo...itale/2007-04-vitale-filmgrain_resolution.pdf
Page 14 says says 35mm format lenses resolve maybe 40 to 140 lp/mm.
And it says Panatomic-X film resolves 170 lp/mm. Color film maybe half of that.
But digital and film are extremely different worlds, different rules.
Nyquist (of Nyquist sampling theorem) Proved that we have to sample AT LEAST at 2x the detail level (basically the line pair thing). But that is the most absolute minimum level of claiming to accomplish reproduction. Yet higher sampling is always better quality of reproduction.
The mistake is to imagine that pixel detail corresponds to image detail in any one for one relationship. Sampling doesn't work that way. We need lots of pixels. Of course, vast areas of our images don't approach whatever maximum we do accomplish. Depth of field sees to that, we are focused at only one distance. But the scene content also contributes to that. We do pretty good at much lower levels.
Your 256 pixels per mm number can at very best resolve 128 line pair per mm, in the most minimal way, 2x sampling is generally deemed less than acceptable quality. So more sampling pixels are always better. At least up to some undefined point that we don't have to worry about. Making this up, but perhaps 2x sampling possibly would be sufficient if we could get all the lines perfectly centered on the pixels, with the same spacing as the pixels, and very straight, not slanted to the pixels?
But the real world is random and chaotic.
I know many are uninterested in any details, they should not read it. But to make the point about how digital sampling works:
Here is an image from a printed Smithsonian Magazine (it being government, I assume I own whatever rights to it that they had). This is a scanner, but a camera samples with the same principles.
I assume it is printed with the normal 150 halftone dots per inch (hard to count). The job is to resolve those 150 dpi dots, a little like line pairs.
EDIT: FWIW, I checked it, and it is 150 dots per inch.
Scanned at 150 dpi - it has the expected moire (aliasing, which is false detail due to insufficient sampling, according to Nyquist):
So here is 2x sampling, 300 dpi (very large, enlarged to match those larger images below, to show the little bump under left eye, at right of where nose was):
At regular size, it looks fairly good, better than above. Here, this is of course image detail, but it is not really an adequate reproduction of the lens detail. We see no round halftone dots. We do know where the edge and little bump is.
300 dpi scans help moire a lot, but for safety factor, I prefer to scan at 600 dpi, and then resample it smaller.
Let's try more, 4x sampling at 600 dpi: (enlarged to same size - starting to get a hint of the dots, but they seem square):
How about 8x sampling, 1200 dpi (enlarged to same size - we actually almost resolve some dots now - with several pixels across the dot to show its round shape. We did not do that before.
Based on this criteria of 8x sampling being better, your 256 pixels per mm can resolve /8 or 32 line pairs per mm of real data. It is not at any wall yet.
Regarding digital sampling, more sampling is usually always a better quality reproduction.
I hate to say it that way, because certainly we can scan resolutions much higher than our goal needs (to copy a photo for example), but if we're going to zoom in so much to examine finest detail (more than we need), then it does show.
But certainly there is no one for one relationship between pixels and line pairs. We always need lots more pixels.
