GF system MTF curves - some reflections

1 month ago

Fujifilm used to publish some MTF curves, like these for the GF 120/4 macro:

There is some issue with them in that they are not physically feasible. The obvious issue is diffraction. No physical lens can achieve MTF = 1.0 above zero lp/mm, due to diffraction.

Very few manufacturers publish MTF data measured from actual lenses. Hasselblad used to do it, Zeiss still do.

Actual measured MTF for a high end Zeiss lens looks like this:

Worthwhile looking at the f/4 curves, 10 lp/mm is around 0.95 while 40 lp/mm is at around 0.87. All figures on axis. (Curves represent 10/20/40 lp/mm)

The Zeiss figure illustrates what real world MTF data looks like. I would figure it is much more credible than the GF 120/4 figure.

But, both have the limitation that they only go to 40 lp/mm. What is the resolution of the GFX sensor? Around 136 lp/mm.

OK, putting that Otus lens on the GFX 100S, what would we expect? Lens rentals has the answer:

Note, red line here is 40 lp/mm

Source

What about the GF 120/4 Macro? We can measure it's MTF from a slanted edge target.

Here is near axis MTF calculated from the DPReview Studio Test shot with the GF 120/4 macro at f/5.6 (blue) , while the red line is my own slanted edge test of my Voigtländer Apo Lanthar Macro 65 at f/4. I have marked the 40 lp/mm and 80/lp mm points approximatel in the graphs. The MTF values are very close, but the GFX 100 has more pixels, so it yields a sharper image

Note also that MTF 50 is around 0.3 cy/px (cycles per pixel) for both lenses. That seems to be typical for high quality lenses.

There is a fundamental difference between slanted edge measurements and measurement on the optical bench. The optical bench measures the sharpness of the lens while slanted edge measures the MTF of the lens, multiplied with the MTF of the sensor and the MTF of the slanted edge. We can compensate for both in the math, which I didn't do here.

A final observation is that both lenses carry a lot of MTF at Nyquist, the vertical lines, any signal right off Nyquist will be turned into false detail.

Putting all the data in a single plot, we could see:

So, what does this tell us? Fujifilm's data is ignoring distortion. MTF measured on the bench ignore the sensor.

Another observation mat be that the 10-40 lp range doesn't tell the whole story, it is far from the pixel level.

Now, lets look at Jim's MTF data for the GF 250/4 at f/5.6. Jim was looking into using the GFX 250/4 with an extender as an long telephoto alternative for the GFX system:

https://blog.kasson.com/gfx-100/how-much-sharpness-do-you-lose-with-the-1-4x-tc-on-the-fuji-250-4/

Two obvious observations are that MTF is near zero at Nyquist . The MTF 50 figure is 0.209 cycles/pixel, indicating a not so great lens.

The GF 250/4 stands high in regard. Is the lens less stellar than it's reputation? Or does Jim have a less than perfect sample?

Would be interesting to know...

Best regards

Erik