Originally posted by ElJamoquio I read you the first time. Both cameras have the same pixel density. Both show a lens that, at f/1.2, is limited by the lens' ability to resolve rather than the density of the pixels.
I disagree. Even in this extreme case (an f/1.2 lens at f1.2) it's not like the border performance of the APS-C is spectacular. At f/2.8, any equivalent-pixel-density FF will beat the APS-C, and at f/5.6 the FF will trounce the APS-C.
When did this become a discussion of equal pixel density? That's not how LW/PH works.
If you want to compare absolute LW/PH numbers across systems (a really bad idea in the first place), you need to know that the numbers scale by
total linear pixels, not by physical pixel density. The denominator "PH" stands for "picture height", meaning the numbers are normalized with regard to sensor size. In practice this means the test chart is shot such that it has the same extent viewed at the same image size. The APS-C test is not a crop of the FF test. This means a 24mp APS-C sensor and 24mp FF will have the same theoretical LW/PH limit. Notice Photozone has the NEX-7 and D3X graphs/numbers scaled very similarly (the difference being the weaker AA filter on the D3X).
Now, because of the nature of using contrast, by way of MTF, to measure resolution, you
can get higher resolution with more pixels even when the MTF50 frequency with the larger pixels does not reach the theoretical limit of the sensor ("lens limited resolution"). Lens blur is not a hard limit, but continuously variable, and the "limit" is determined by how you sample and measure it. Pixels can only assume discrete values, and smaller pixels can sample and distinguish finer gradients of contrast than large ones. This can easily increase the frequency at which the contrast level falls below the MTF50 contrast threshold without increasing lens resolution.
Take a look at these two tests of the same lens with the same size APS-C sensors, the only difference being one is 10mp the other 16mp:
Nikkor AF-S DX 35mm f/1.8 G - Review / Test Report - Analysis Nikkor AF-S DX 35mm f/1.8 G - Review / Test Report - Analysis
Notice the curves are almost exactly identical, and at f/1.8 are well under the resolution potential of both sensors. Notice how moving from 10mp to 16mp the LW/PH figures of both the "border" and "extreme" are scaled very closely to sqrt(16/10), leaving the overall curve, and ratio of resolution to theoretical maximum, relatively unchanged (note, though, that this scaling is not linear on larger scales as it is limited by diffraction and particular characteristics of the lens' PSF of course).
Now back to the Canon 50 f/1.2. Here is Lenstip's test, shot with both the 50D (15mp) and 1ds Mk III (21mp):
Canon EF 50 mm f/1.2L USM review - Image resolution - Lenstip.com
Lenstip quantifies resolution in lp/mm (line pairs per millimeter), which really contains the same information as LW/PH (as long as you know sensor size), only it is resolution density rather than total lines. The 50D's border numbers exceed the 1Ds3's at all apertures, even when factoring in the different "decency levels" Lenstip assigns to each sensor. And what's really interesting is the comparison between the 50D and the "APS-C edge" on the 1Ds3's graph. Notice that the 50D's curve has the same shape, but the numbers exceed the APS-C edge on the 1Ds3's at all apertures, demonstrating my point that increasing sensor resolution
does matter even when "lens resolution limited".
Anyway the
real moral of this story is that pictures shot on APS-C and FF with the same lens are not comparable! Comparisons of LW/PH or resolution "sweet spots" are meaningless, because the images are totally different! And then there's the fact that lens resolution can only be measured as part of total system performance (including RAW/JPEG image development), and not in isolation. There's a reason why Photozone has a disclaimer that results are not comparable across systems.