On Tue, Oct 17, 2006 at 09:41:54AM -0700, Shel Belinkoff wrote:
> Hi Gang,
>
> I've tried following the various threads and messages about the Pentax
> Nucor 22-bit converter-chip-engine-whatever-it-is thing, and have either
> been lost in the dust of techno-speak and the more advanced discussions, or
> turned of by the idle speculation of one thing or another. However, I've
> been trying to understand what's going on inside the K10D.
>
> So, here's the way I see the new Pentax system. It's using the 22-bit Nucor
> technology to produce a 12-bit output, but because the processing starts
> with a far greater input, tonality, detail, smoothness, etc., is improved
> (somewhat akin to working in 16-bit mode in PS and then converting to 8-bit
> for printing or web work). At some point, it seems, the output can be
> boosted to 14-bit or 16-bit. The file sizes would be quite a bit larger,
> however, with all the problems that involves.
>
> Speculation: If the response to this technology is good, Pentax may be able
> to boost the output to 14 or 16 bits on subsequent cameras, such as the new
> higher spec model they are rumored to be working on, or an upgraded version
> of the K10D (K10D-II).
>
> Does my understanding of the technology seem about right? If not, in plain
> ol' simple English, where does it fall down.
It's more or less correct, overall. It falls down on a couple of things:
o The assumption that the process starts with a far greater input.
In fact it's very much like working with 16-bit mode in PS;
that's worthwhile, even when starting with only a 12-bit RAW file,
to avoid introducing any additional errors during the processing.
But just as working with 16 bits in PS can't create any new data,
the processing in the DSP is limited by the signal quality of the
input (the sensor, in this case).
o The acceptance of the implicit claim that 22-bit is better than
18-bit (or even 16-bit, until sensor quality improves dramatically).
Nobody is saying that pre-processing is a bad thing, especially if
some kind of gamma correction or tone curve is applied to more
closely match the 12-bit quantized output space to human perception.
And if the most suitable processor for the job happens to have a
22-bit data path, then why not use it? But in reality you'd get
just as good a result with an 18-bit processor (if one existed).
If, as has been suggested, Pentax plan to switch to 14-bit (or
even 16-bit) output the benefits of the pre-processing decrease;
I'd expect results produced (in PS or the like) starting from a
set of 16-bit un-processed sensor values to be indistinguishable
from those produced from values pre-processed in the camera (for
all practical purposes; pixel-peeping to show that in one case
the pixel value is 3587, while the other method produces 3586,
doesn't interest me).
--
PDML Pentax-Discuss Mail List
[email protected]
http://pdml.net/mailman/listinfo/pdml_pdml.net
