Bernhard Ess said:
I dont understand the technical side of it, but I am surprised
because the USAF targets are widely used for testing scanner
and camera resolution.
It's a bit like newspapers. Just because they are widely distributed,
doesn't make the information necessarily true or unambiguous.
Can you say how resolution numbers based on sinusoidal/ noise
targets will typically behave compared to current resolution
tests? Can they be compared directly? Will they tend to indicate
more or less resolution?
Several things make a direct comparison difficult or even impossible.
First, imagine one of those very fine (bi-tonal) line patterns to have
exactly the same spacing as the sampling density of the CCD sensors.
It is then theoretically possible to have an exact alignment and the
target will be accurately resolved, but when the same target is
displaced by half a pixel the result is uniform gray (totally not
resolved). So target alignment (even more complicated by slight
rotation) will have significant effect on the observed/apparent
resolution, especially if the scanner optics are good.
Second, the sharp edges of the bi-tonal bar patterns will cause
aliasing artifacts. That will cause lower spatial frequency patterns
(seemingly wider bars) to mix with the real pattern. That can make it
more difficult to get to the real resolution. It can result in local
seemingly higher (or lower contrast) which has nothing to do with real
resolution.
Third, those bi-tonal bar patterns have a certain light/dark contrast.
Higher (or more importantly lower) contrast will result in different
outcomes for resolution. It is impossible to estimate the influence of
contrast on observed resolution with such a target.
The MTF measurement methods on the other hand, take care of all those
uncertainties. It is now even very simple to establish those MTF
results in a uniform procedure, with the public release of "Imatest"
(
http://www.imatest.com) which uses the "slanted edge" method.
Bart
