> Thats not true. Nikon and Canon both do a lot of their own sensor to
> digital representation processing too, ya know? And their RAW data is
> just as much RAW as anyone else's. I don't know what quantization
> they use in their A-D converter; I suspect it's greater than 12 bits
> at least in the higher end cameras ... I suspect 14 or 16 bit, it
> would have to be if they're concerned with round-off error in
> modeling the voltages for a good 12 bit representation. Pentax has
> gone beyond that with a 22 bit A-D, which makes much more accurate 12
> bit output. 22 bit quantization vs other cameras' 12, 14 or 16 bit
> quantization renders better modeling of the voltages regardless of
> whether the output digital representation is further reduced to 12
> bit or not.
>
Umm... sorry. I'll have to throw the "bullshit flag" on that
play. If one measures and analog voltage at 22-bits, does nothing to it,
and then throws away the least-significant 10 bits to produce 12-bits, the
results are absolutely identical to another who quantizes at 12-bits to
begin with. Only if "mid-processing" (i.e. processing between the data
measurement and RAW file writing) is done is there a difference. That
"mid-processing" is precisely the unknown that I'm talking about, but one
would assume that some was done.
Most likely the reason for using 22-bits was so that variable-gain
analog amplifiers and fixed-full-scale-voltage A/D's were NOT used. For
ISO 100, choose bits 10-22. For ISO 200 choose bits 9-21. ISO 400 choose
8-20, etc, etc.
Now, for example, if one were to apply a cloudy WB vs. a tungsten
WB, the red channel gain would likely be at least a bit (i.e. f-stop) or
two different. It makes sense to chose a different "dynamic range" for
each channel based on things like WB (and by association colorspace).
> I don't presume to know what Pentax did in tuning the A->D converter
> and subsequent 22->12 bit transformation, but whatever the body
> writes to the RAW format file as sensor date is, by definition, the
> camera's RAW format data.
>
Yes, but no longer presumed to be an unaltered digitized version
of exactly what the sensor sees.
> You can't get at or use anything more anyway so there's no point in
> debating it.
>
If you get the white-balance wrong, you only get 10 or 11 (linear)
bits of the red channel in the example I proposed above. That does make a
difference.
For the most part, this is speculation since nobody but the
image-processing software guys at Pentax know for sure. I'm just saying
that not only is what I'm suggesting possible, it's rather *likely* given
that they're trying to exploit the additional data available.
-Cory
--
*************************************************************************
* Cory Papenfuss, Ph.D., PPSEL-IA *
* Electrical Engineering *
* Virginia Polytechnic Institute and State University *
*************************************************************************
--
PDML Pentax-Discuss Mail List
PDML@pdml.net
http://pdml.net/mailman/listinfo/pdml_pdml.net
