I repeated my experiment at f/22, but omitted capture sharpening. The resolution calculated from your target was 57 lp/mm and the MTF 10 calculated by Imatest was 59 lp/mm, confirming the approx. 10% figure you quoted. It is interesting to note that the MTF10 at f/22 is at Nyquist, which would imply that deconvolution sharpening could restore some of the lost detail imposed by shooting at this small aperture.
That's correct. Unfortunately we'll not be able to recover all detail all the way up to Nyquist in your example because the actual image contrast gets too low (in case of an adequate Low-pass filter). Image contrast multiplied by MTF response approaches zero. This is also where a low glare lens helps to maintain contrast in the microdetail. I expect the image contrast of a MF system without low-pass filter to reach Nyquist without much attenuation other than that from an area sample (and produce aliasing).
I am somewhat confused by your statement that MTF at Rayleigh is around 20%, since I had always thought that it was aroung 10% as stated by Roger Clark. However, on checking in Wikipedia, they do mention a figure of 20%. How does one resolve these conflicting values?
Frankly, I traditionally assumed Roger's 9% number to be correct, but when checking Wikipedia found a higher contrast being quoted. To verify, I made a simulation in ImageJ of two (f/16, 564 nm) diffraction patterns I already had, offset by the radius to the first zero, and I find something like a bit over 26% contrast (depending on sensel size (!) and alignment with the sensel grid). This simulation is without the effect of a low-pas filter.
P.S. The sensel's aperture size (or microlens) will quickly reduce the above oversampled (1 micron) rendering of the neighboring diffraction patterns (Rayleigh criterion) into a complete blur with zero contrast. I'm preparing a demonstration of it, but it takes the computer a while to calculate the base images.