Before I deal with all the uncontrollable and semi-controllable variables involved in making actual test photographs, I did some simulation to make sure the target would work the way I thought it would under completely controlled, albeit unreal, conditions.
I wrote this Matlab script:
I ran it, and plotted the results.
As expected, more low-pass filtering reduces the standard deviation:
The standard deviations are remarkably insensitive to resampling, both with and without an antialiasing filter:
And finally, cropping random, same-sized portions out of the image doesn’t change the standard deviation much (note the vertical axis scale):
Next, I need to print the target and make some images.