Re: Corner drop off
Hi Jerome,
I would love to agree with you, but I can't.
Why I can't? One reason is that our vendors don't give us data for high frequencies. Hasselblad, Zeiss and Leica all use the frequencies 10/20/40 lp/mm. Rodenstock and Schneider have some MTF data at 15/30/60 lp/mm. That may be a bit better than the usual 10/20/40 lp/mm but is not even close to the 97 lp/mm the IQ-180 has, just as an example.
More importantly, it is long established knowledge that low frequencies dominate human vision. So low frequency MTF is far more important than high frequency MTF. Also high MTF at resolution limit is not really desirable as it causes aliasing artefacts. Common knowledge…, would you find my remark ignorant, here is a source:
https://en.wikipedia.org/wiki/Aliasing
Here is a paper by Zeiss on interpreting MTF curves, a lot of good wisdom:
https://www.zeiss.com/content/dam/Photography/new/pdf/en/cln_archiv/cln30_en_web_special_mtf_01.pdf
Perceived quality can be measured by SQF, a value based on integrating the MTF over the contrast sensitivity curve of human vision. You find a good discussion of the concept around page 7 in this document from Zeiss:
http://www.zeiss.com/content/dam/Photography/new/pdf/en/cln_archiv/cln31_en_web_special_mtf_02.pdf
Now, that I have suggested some litterature for insights, here is the real reason I would not agree with your point:
In real world, we either see images as prints or on screen at small resolution. In neither case can we really see pixel level detail. In print, the fine detail is normally at a size where the CSF (Contrast Sensitivity Function) of human vision is low. On screen size images most of the detail is thrown away anyway.
If you make a very large print and view very close, we would be able to see high frequency detail. Lens MTF data doesn't say much about that region, that applies to lenses from Hasselblad, Leica and Zeiss. Schneider and Rodenstock gives a lot more data.
You can measure MTF using Imatest. I do it often, but not in the corners. Why I don't do that? Because it needs a very large target. Ideally something like 50x focal length is needed, but also because I am not so terribly interested in corners.
In the case I have shown, the Zeiss Distagon performs better in the extreme corners but the Canon performs better at the edges and just a bit in from the corners. Most of the older Zeiss designs drop of significantly at the last few millimetres of the diagonal. The newer lenses are designed for a larger image circle, it seems.
Best regards
Erik
Were the other images any less real?
Not only in the corner. The 2/3 zone is critical, especially for wide-angle and zoom lenses.
But you are still using MTF curves with values of 10, 20 and 40 cycles/mm, which would be adequate for 12 Mpixels sensors. For the A7RII, we would need curves at 100 cycles/mm, wouldn't we?
- - - Updated - - -
Your wish is my command: