danke klaus, aber...

So I've been trying to understand why, when I compare my own tests of lenses I own from the list of lenses tested on PhotoZone, I find his MTF50 numbers do not correspond to my perception of sharpness, but rather something closer to resolution.

I can see two possible explanations of this, which probably both contribute to the mismatch.

First, as Klaus mentions in his Lens Test FAQ, Imatest tests "sharpness" but not "contrast". As Klaus himself admits on his MTF page though, human perception of sharpness has at least as much to do with low-frequency contrast, as with edge sharpness (acutance). It'd be fine if it was made more explicit that MTF50 was somewhere between contrast and resolution, without necessarily being indicative of either. As it is, his results show low-sharpness lenses such as the kit lens with "sharpness" numbers approaching those of L primes. The newest digital zoom lenses all have MTF50 numbers close to the theoretical maximum for Canon cameras, so it is hard for better lenses (like primes and some of the L zooms) to show much numerical improvement. I'd be more interested in MTF70 or MTF80 numbers.

Klaus asserts there is no way to test lens contrast. This seems untrue: even if it is difficult to get absolute numbers for contrast (Photodo claims to be able to do it), it is trivial to compare lenses with the same camera mount. Just use the same setup, and swap the lenses. I do it at home, and the difference is obvious when comparing the resulting histograms (which are just sets of numbers after all).

The second explanation: Klaus is testing on oversharpened images. They may be fine for print, but pixellization and sharpening halos are much stronger in Photozone images than sample images from the other major camera review websites. If you think I'm just oversensitive to sharpening, or assume he used less sharpening for the test, consider that DPReview also tested the combination of the EOS 350D and EF 50mm f/1.4, and came up with a resolution (defined as the spatial frequency at which details can just barely be differentiated) much lower than PhotoZone's MTF50 number (at which frequency, details should still be quite obviously differentiated). The DPReview test shots appear to use standard in-camera sharpening.

BTW, this oversharpening is why you should not compare lenses based on JPEGs taken under uncontrolled conditions, unless you are really familiar with how to equalize sharpening across images. This is why Imatest provides the option of "standardized sharpening" on its tests.

Of course, this is all conjecture, because I can't recreate PhotoZone's test conditions exactly, and I don't have Imatest to confirm that the results are affected. And PhotoZone's results are not that far from the reality I see - it's just that the reality diverges more and more with the new digital lenses that appear to be pushing the resolution envelope at the cost of contrast and color. Sharpening less would decrease the extent to which lenses of this type are favored by this testing methodolgy.