Rob Geraghty wrote:
> When I was reading something someone else wrote on this topic I couldn't
> help wondering about the kind of oversampling used in CD players to filter
> the output. I wonder if similar technology could be used to smooth the
> output from a scanner - maybe some scanners already do? Oversampling in
> a CD player just interpolates points to cheat the nyquist limit. Presumably
> in some way this is analogous to what "true fractals" does?
I've been steering clear of this thread, but I think this question is clear cut
enough to avoid further confusion.
Probably famous last words time, - but here goes.
Oversampling in audio is a hardware implemented function, which is really only
of any worth when it's applied at the recording stage. The primary use is to
make the 'brick wall' low-pass filter more effective (and cheaper to make) prior
to the A to D conversion stage.
The analogue in scanning would be to sample at a higher rate, ie. more pixels
per inch, and then to resample to a lower resolution before taking the image
information out of the scanner. That way you'd have manageable file sizes, and
hopefully, reduced aliasing artefacts.
Any manipulation done after the scan would have to be *much* cleverer than
simple interpolation to clean up an already aliased signal. Cleverer than
'Genuine Fractals' even.
A researcher at Microsoft (is there nothing they won't stick their clumsy
fingers in?) has shown, crudely, how oversampling can be used to reduce aliasing
in digital HDTV without increasing the bandwidth to accomodate high sampling
rates, or requiring spatial low-pass (or should that be long-pass?) filtering.
To get back to applying oversampling to scanners, there are plenty of 10,000
pixel per line linear CCD arrays around. It wouldn't take much to combine an
area of 4 samples into one, using a true RMS algorithm. This would mean that the
scanner would output only 5,000 dpi, but the brightness level could be much more
accurately represented than simple area integration allows.
Bi-cubic interpolation of a 16 pixel area would probably be better, but we don't
want to get ahead of ourselves, do we?
BTW. Does anyone have the details of Kodak's 'accurately fuzzy' optical filter,
or does it simply exist in an accurately fuzzy description?
Regards, Pete.
