After combing through ISO 12233 and reading multiple articles, I'm still unsure what LP/PH values are considered poor, good, or great. I just finished testing a Nikon D750 with an 18-70mm lens. I'm using Imatest to do the analysis. For example, I'm analyzing 9 regions of the sensor, and I'm getting values ranging from 486 LP/PH in the bottom left corner to 1261 in the center.
According to this page:

Lens MTF Test Info

30 lp/mm > 0.5 is very sharp 30 lp/mm > 0.3 is sharp if you sharpen a bit 30 lp/mm < 0.2 is getting soft

For a DX lens, the Picture Height (PH) is 16mm. This results in (30LP/mm)*(16mm/PH) = 480 LP/PH. This would suggest if MTF50 > 480 LP/PH is sharp.

However, on another forum, I saw a comment that said:

I find 1400LP/PH acceptable, but 2400+ preferable.

If 480 is very sharp, why would someone fine 1400 only acceptable?

Any assistance, reference, etc. would be .beneficial

  • \$\begingroup\$ If you aren't taking some very meticulous measures to insure your perfectly flat test chart and your perfectly flat camera sensor are perfectly parallel to one another, any results using Imatest will be totally bogus. \$\endgroup\$
    – Michael C
    Commented Apr 5, 2021 at 7:56
  • \$\begingroup\$ I have a nice and repeatable studio setup. The target is flat and vertical. The illumination is measured to be uniform within 10%. I wrote a Python script to compute the eSFR based on the algorithm described in ISO 12233:2017 and the results were consistent with that of Imatest. I am confident of my setup. Now, I just need to understand the results. I mentioned the random person as an example of growing information I'm finding during my research. \$\endgroup\$ Commented Apr 6, 2021 at 14:43
  • \$\begingroup\$ It doesn't matter how perfectly flat and vertical the target is. If it is not more or less perfectly parallel to the camera's sensor you will get skewed results that will be indistinguishable from lens tilt. That can explain extremely poor results in one or two (opposing) corners. As for the internet comment, unless you (and if you do, please share it with us) know the context of the format size being discussed, it's pretty much meaningless. One should expect much higher lp/ih numbers for larger formats than for smaller formats. \$\endgroup\$
    – Michael C
    Commented Apr 7, 2021 at 5:11

1 Answer 1


MTF is a system measurement; it is not a lens measurement. So, what is "good" depends on the capabilities of the system as a whole and the combination of limitations. E.g. lack of lens sharpness due to spherical aberrations, image softening due to AA filter, and max sensor resolution.

I'm assuming a typo and you are referring to the DX D7500. That has a vertical resolution of 3712px, max 1856 lp/ph, 116 lp/mm. Of course, that is theoretical, based solely on sensor pixels, and would be at a very low contrast. But, it is not even theoretically possible for your camera to reach 2400 lp/ph; so applying that standard makes no sense. Also note that you may see Imatest results in l/ph... given that measure the D7500 could theoretically reach 3712 l/ph. And people mix up their units of measure all the time (or they are not clearly stated).

The very best lenses reach over 80 lp/mm on a very high resolution sensor w/o AA filter (e.g. Nikkor Z 85/1.8 on a Z7). But, in general, 30 lp/mm is considered a good level of decency. Why 30?

What 30 lp/mm means is ~ 1.4 MP recorded on a DX sensor... that probably sounds crazy low; but the CoC standard for image sharpness is .02mm for a 1.5 DX sensor. Which equates to 50 l/mm or 25 lp/mm (the accepted standard requires less than 1MP recorded on any size sensor). And 30 lp/mm exceeds the 25 lp/mm standard.
But that 1.4 MP result is also an MTF50 number, and there is almost certainly a lot more resolution at lower levels of contrast which can be improved (sharpening is just adding contrast). And it's not like the typical human's vision just fails below 50% contrast.

Nikkor Z 85/1.8 Z7 test in lp/mm:
Nikkor Z 85/1.8 Z7 test in l/ph:

  • \$\begingroup\$ Thank you. Your response provided a great deal to digest. I was referring to my full frame D750. When I put the 18-70 DX lens on it, it cropped down the image. The assumption I made in my post was that only an area of (16mmx24mm) was used of the available (24mmx36mm) due to the lens. This is why I used a picture height of 16mm in my calculation. Also, I do realize that I am unable to separate the components of the test (quality of print, camera, and lens) however, the setup provides a quantitative basis to compare all my cameras and lenses and test for flawed equipment. \$\endgroup\$ Commented Apr 6, 2021 at 14:58
  • \$\begingroup\$ Since writing the original post, I decided to use the MTF%>.9 for 10 LP/mm as a measure of high contrast and MTF%>.5 for 30 LP/mm as a measure of sharpness. This is implemented by taking the spatial frequency response and finding the LP/PH values corresponding to .9 and .5. I then compare those values to (10 LP/mm*24mm==>240 LP/PH for a full-frame sensor) for high contrast and (30 LP/mm*24==>720 LP/PH for a full-frame) for sharpness. So far, the results using combinations of D750, D500, Z6, 300mm 2.8, 70-200 2.8, 18-70mm DX are consistent with expectations. \$\endgroup\$ Commented Apr 6, 2021 at 15:13
  • \$\begingroup\$ @AdrianHoodSr, I didn't recognize that the lens was DX and forced a DX crop; but makes sense. Your new standards are more in-line with the human's optical response curve and similar to the methodology DXO uses to determine "perceptual MP's." \$\endgroup\$ Commented Apr 6, 2021 at 15:31
  • \$\begingroup\$ lp/mm and l/mm is basically the same thing. Line pairs per mm counts both the black and white lines. Lines per mm only counts the black lines. \$\endgroup\$
    – Michael C
    Commented Apr 7, 2021 at 5:16
  • \$\begingroup\$ @MichaelC, no. l/mm counts both black and white lines and is 2x the lp/mm value. image-engineering.de/library/technotes/… \$\endgroup\$ Commented Apr 7, 2021 at 13:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.