Jan@xxxxxxxxxxxxxx writes:
<< But I could be wrong. Maybe someone who has studied signal processing and
information theory closer than 25 years ago has better insight.
Tim? >>
Jan,
I am not sure I have much better insight but here are a few thoughts
with a slightly different perspective and some speculation:
<< By moving cells (or moving your source material) an increment of the cell
size, you in effect average half of two adjacent cells, but the
result is no different than if you did it in software, except slower.
>>
1) In mechanically averaging adjacent pixels (cells) by jittering the CCD
sensor (probably ideally in a circular pattern) , there can be some advantage
over digitally averaging from a fixed sensor. The mechanical averaging takes
place before digitization. This has the effect of low pass filtering before
digitization and hence eliminates/reduces aliasing. (search archives for
previous discussion/description on aliasing). Aliasing tends to introduce
"noise" artifacts where there is fine detail of a similar size to the pixels
and also intoduces moire type fringes where there are repetitive fine details
in the scene.
<>
2) In packing the octagonal cells in a more symmetrical manner the resolution
becomes more independent of orientation. In a square cell the theoretical (un
achievable) maximum resolution is at most the pixel spacing in the horizontal
direction. Similarly for the resolution in the vertical direction, if the
spacing is the same vertically. However at 45 degrees the resolution is only
about sqrt(2 ) ~ 70 0f that in the horizontal direction. So the resolution
should be more independent of orientation with octagonal pixels. You would
need to know the exact packing arrangement to analyze this fully.
Interestingly you might be able to achieve something similar to (1) by just
using a lower quality lens as the blurring will average between adjacent
pixels automatically! I would guess the lens should ideally have an aperture
chosen so the main lobe of the lenses "airy disk" (sin X/X impulse response)
overlaps two adjacent pixels. This would maximise performance. This is a case
where an inferior lens might give improved results. I have always wondered
whether the grain structure of film might not interact with a really good
enlarging lens to give degraded results, since the grain acts to digitize the
picture to at least some degree.
Probably mechanically averaging is still better than optically
defocussing/blurring. There is probably an optimum jittering pattern
dependent on the overall lens resolution and pixel spacing : this translates
to a weighting function on areas of adjacent pixels. Often these sorts of
problems have optimal solutions that look like si(x)/x functions.
Regards,
>>Tim Hughes<<
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|