It's difficult to estimate how much advantage there is to having 16
bits of quantization space vs 12 bits without more data, with respect
to total dynamic range in stops or EV numbers.
The analog capture range of the sensor from saturation input to noise
threshold ("how much energy will max a photosite voltage rise to how
little energy will trigger a photosite voltage rise") is one measure
you have to know. You also have to take into account the typical
gamma curve used to transform the linear capture space of the sensor
to a rendered RGB image. The gamma curve compresses together the high
values and spaces apart the low values to fit the captured data to
the appropriate brightness range required for human vision. You also
need a measure for 'acceptable noise' at the black point clipping
level. Let's presume this latter is a constant, whatever it is.
With todays 12bit sensors, quantization depth would net 12 stops
tonal capture theoretically, but post-gamma correction the resulting
output is in the range of 7-9 stops maximum, given the analog dynamic
range limits of the sensor.
If the K10D sensor has the same analog range of sensitivity as the
current 6Mpixel sensor and 16 bit quantization space instead of the
current 12 bit, what this means is that it can distinguish 16x more
tonal steps in the analog range of its linear capture space.
Processing this input with a 22bit image processing engine poses an
advantage in reduced round-off error and accurate representation of
the captured data into final storage form. If the analog dynamic
range of the sensor is the same, it will still have the same 7 to 9
stops of dynamic range, but they will be more accurately represented.
That's as much as we can say without knowing the analog dynamic range
of the sensor in question. Medium format backs with 16bit sensors
typically have additional analog dynamic range in addition to larger
quantization space, netting an increase in output dynamic range up to
the 12 stop range with more accurate representation of tonal values
captured ... there's a reason these MF sensors are expensive both in
price as well as space and power requirements.
BTW: 12 stops of analog dynamic range surpasses any film I've ever used.
Godfrey
On Sep 6, 2006, at 12:52 AM, Jostein Øksne wrote:
> With all the talk of a 22 bit A/D converter for the K10D, and the
> corresponding speculations of true 16-bit colour depth in the
> raw-files, there are a couple of things I wonder about.
>
> Firstly, I wondered what the competition was doing. In the 35mm realm,
> Canon use 12-bit colour depth in both 5D and 1DSmkII. I didn't check
> other models. Leica, however, use 16-bit for the R-series digital
> back. In the medium format realm, it seems that all the makers except
> Mamiya use 16-bit. I have checked Sinar, Leaf, PhaseOne and Imacon.
> Mamiya ZD use 12-bit colour depth, but 14-bit A/D. I couldn't find any
> info on the A/D conversion for the other brands.
>
> Anyway, it seems that 12-bit is a standard for 35mm, and that 16-bit
> colour rule the medium format world.
>
> Better dynamic range has been mentioned frequently in the discussion
> of 16-bit colour, but I don't think I paid enough attention. How much
> would 16-bit depth improve the dynamic range over 12-bit in terms of
> f-stops? Is this going to be like going from slide film to colour
> negatives, or is this on a different scale?
>
>
> thanks,
>
> Jostein
>
> --
> PDML Pentax-Discuss Mail List
> [email protected]
> http://pdml.net/mailman/listinfo/pdml_pdml.net
--
PDML Pentax-Discuss Mail List
[email protected]
http://pdml.net/mailman/listinfo/pdml_pdml.net
