[Openicc] beyond 8-bit precission [was: new version of xcalib]

[Graeme Gill]

Kai-Uwe Behrmann wrote:
>>But that is not actually a major limitation, if you are careful
>>in your color configuration. The input end (8 bit per component)
>>is a colorspace with a transfer curve that you are defining with
>>the RAMDAC values. It's therefore possible to choose the values
>>of your 256 steps so that each step is visually imperceptible.
> 
> 
> Seems not possible. I can do what I want 8bit steps are allways visible on 
> my LCD (average model). An synthetic gradient shows clearly what I mean. 

It depends how you are setting up your 8 bits space. The claim is
that 8 bits is just enough under the right circumstances. There is
little or no margin though.

> No black/white photographer is convinced by an maximum of 8-bit level of 
> sensitivity.

But if they are dealing with device space (ie. photographic film),
or capturing a high dynamic range and expecting to extract a
"finished" image from it, then of course 8 bits will not
be enough. 8 bits is just enough for a final image, if it
is represented in the right colorspace.

 > In medical practice 8bit is not used for the same reason
> (X-ray images).

Once again, this is a device space (film), and it's intentionally
recording a high dynamic range (radiographers want to be able
to examine different density sections of the image), so of
course 8 bits is insufficient.

>>(i.e. back of the envelope calculation shows that a 0 to 100 L*
>>range divided equally into 256 steps gives steps of 0.39 delta E,
>>less than half of the perceptible threshold.)
> 
> 
> This calculation refers to an reflective media, right?

No, to any image that has been processed and is in a final
state, so that it has a black point down to L* = 0, and
a white point of L*=100. Most computer display devices (CRT's,
LCD's etc.) conform to this assumption. Other technologies
don't (ie. film projection), where there is the possibility of
"white than white" highlights.

> Todays inkjets have left behind 256 steps - steps in the sense of 
> stepping. Tell me if you like, I can send you an print, which shows 
> clearly each step of 8-bit shades of gray output.

Once again, I was referring to a carefully constructed space, not
a device space. The whole point of looking for RAMDAC tables with
greater than 8 bit entries is to handle the requirements of device
spaces with sufficient precision to be able to control them.

As for inkjets, we were using 8 bit/component systems extremely
successfully in commercial high quality proofing, and the 8 bit
limitation was never an issue. The reason was simple, the
system has device calibration, and the 8 bit space was carefully
constructed to have a close to perceptually uniform transfer
characteristic. The output of the calibration LUTs (device space
values) was 12 bits/component.

> This calculation refers to an reflective media, right?
> Todays inkjets have left behind 256 steps - steps in the sense of 
> stepping. Tell me if you like, I can send you an print, which shows 
> clearly each step of 8-bit shades of gray output.
> 
> Todays artists need more precission!

For creative use (where images are going to be edited, composed,
or stored in device spaces), then I fully agree that this is very
desirable. In practice, there are many 8 bit/component systems
and storage formats out there, and we have to make the best of them.
If they are used for final images in an appropriate space, the 8 bit
limitation need not result in visible artefacts.

Graeme Gill.