to resolve 1 line you need 2 pixels,
Only if you want to resolve those lines poorly. It takes 5 to 6
pixels to properly render a line pair.
John, this is not true for sine waves and when we talk about MTF
versus frequency we are talking about sine waves. The examples I have
seen that claimed to show 5 to 6 oversampling is required either used
poor reconstruction techniques or they were imaging sharp edged lines
that required not just the fundamental but also the third harmonic to
maintain the sharp edges. For a sine wave you only need a little more
than the Nyquest limit: perhaps 2.5 times the frequency. On the other
hand Bayer arrays only have unambiguous luminance information for
half the pixels so for a Bayer sensor you need to increase the pixel
count by a factor of two compared to a monochrome sensor.
I want to see the optics without artifact; not count sine waves when
they register with luck of alignment.
I think the whole Nyquist thing is poorly thought out with no
foresight; 2x the highest frequency is not sufficient for sampling;
you no signal with just the "right" (wrong) phase, and you get
amplitude modulation with simple integer ratios. The Nyquist is
insufficient, from a high-quality perspective. It is "good enough"
for crude purposes, and with the understanding that the highest
frequencies are ignorable in the sampling system.
Try it for yourself; make a sinusoidal image of alternating
white-grey-black-gray-white across the image with a period of about
36 pixels. Then, do a very slight perspective correction so the
frequency is a little different at the top and bottom. Open a
pixelate filter in preview mode, and try various numbers. I think
you'll see that the signal is not artifact free unless the tiles are
6 pixels square or less. At the nyquist (18x18 tiles for a period of
36 pixels), what you get can be best described as "garbage". This is
much more obvious with graphics than with audio.
--
John
)'; 
