>
> >They don't dither? They should - on both upconversion and downconversion.
>
> Dithering is completely different from re-sampling. Dithering is the
> mixture of small dots of different colors to create the perception of
> a third color. It is generally a separate process that is applied to
> an image, most notably in print drivers or RIP software. Some screen
> drivers (such as QuickTime or PDF viewer) employ dithering if they
> know they're displaying on a low-bit (256 color) display.
That is one application of dithering, but it is not the definition
of dithering. Dithering is the addition of a noise signal to
improve the statistical properties of sample conversions. In the
case you mention it is for a drastic downconversion. (BTW, for the
actual applications you mention, dithering is frequently faked. You
can see tiny blocks with identical "dither" patterns.) Dither is
also used for less drastic conversions. You mentioned expanding
color ranges. For example if red has 8 bits, but a picture only
uses values 0-204. To expand to 0-255, you shouldn't just multiply
by (255/204), then round. No input value can convert to 187, and
there will be 50 other gaps in the probability distribution
function. A tiny bit of properly-shaped dither added to the input
will result in a output distribution that actually looks more like
the input distribution. Dither helps with both up- and
downconversion.
>
> All "standard" re-sampling methods, such as nearest neighbor,
> bi-linear, and bi-cubic, do not employ dithering.
Many would perform better with it, if they are not doing it without
your knowing it. Not doing dither properly does seem to be very
common. I think it's partly due to having software engineers
writing their own signal processing routines. (No affront intended
to software engineers. It's just that they usually haven't been
trained in this.)
>
> --
> : Jan Steinman <mailto:Jan@xxxxxxxxxxxxxx>
< 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 >
|