PS Uses Relative Colorimetric Intent with BPC for Soft Proofing? - Adobe Color Management

Kwanyue & Gernot,





Thanks for revisiting the topic of monitor blackpoints.





I think that D.7.8 just says that, since most users expect R=G=B=0 to
produce max black on the monitor, most users would also want BPC in the
conversion from PCS to monitor space (so that shadows are never clipped).>
Whether or not one could see the effect of BPC would, I should think,
depend on the the level of black that your monitor can achieve: the higher
the blackpoint, the more noticeable. Of course, the monitor profile would
have to have a blackpoint tag or equivalent, and Photoshop would need
to honor it.




I believe BPC in softproofing is just a unique PhS feature to make blacks looks richer and fuller on monitor as ANNEX D7.8 states:

Furthermore, the monitor profile transforms that are common on many systems are based on oversimplified mathematical models. Often they take the form of a linear transformation from XYZ to RGB (a 3 x 3 matrix) followed by a simple power law in each channel for gamma correction. Such transforms often fail to model the behavior of the monitor accurately in the shadows, since they ignore the biases that commonly occur in the CRT and support electronics. These biases are variable from unit to unit and are also dependent on the user selectable settings of contrast and brightness. Fortunately, any departures from colorimetric accuracy that result from these simple models are relatively minor and are partially masked by face-plate reflections, often 3 to 5 percent, so that they are generally tolerated.



For Y=0.002121, I'm not sure you'll notice any outright clipping even
if playing with a pure gradient. You might notice a slight fogging up
of the scene when switching on BPC (with RelCol), due to the linear rescaling
of all scene luminances upward to fit inside the monitor bp/wp range.




To my own knowledge and understanding most monitor profiles do not have BP tags and PhS do not recognize BP tags in monitor profiles in softproofing and BP to be zero in luminance and chroma so the fogging up should be explain by above paragraph in ANNEX 7.8. It do however recognize BP tags in profiles when doing color transform eg. sRGB to CMYK profile if the profiles contains BP tags.

Suppose Photoshop does honor version 4 profile bp tags, but you don't
see much change, then I would try to bump the bp value up (maybe manually
edit the tag with a binary editor) to see if this does anything. That's
what I would do anyway if I have the time (and a profile with a nonzero
bp).





Regards,





Larry

Gernot & Kwanyue,

Thanks for your comments.

If, in Gernot’s point #2, the working space is monitor space, with the RGB values sent directly to the monitor without further intervention by the CMM, then there would be no clipping, assuming monotonic curves, etc. However, it seems that the problem could occur in the conversion *to* monitor space under the following scenerio:

Suppose the CMM, for whatever reason, uses absolute black as the black luminance for it’s BPC calculations when in reality the monitor’s black luminance is actually not that black, then the CMM miscalculates when BPC is ON, and makes all scene luminances darker than they should be. The deepest shadows should clip in such a case, don’t you think?

Larry

Adobe black point compensation:
<http://tinyurl.com/j52xy>

Larry,

a mathematical model might be helpful:

x Working space data, linear RGB or Y, 0..1
y Data, as sent via graphics card to the monitor, 0..1
z Measurable monitor luminance, 0..1

y(x) = f(x)
This is a so far unknown function

z(y) = b+(1-b)y^G
TRC for the calibrated (!) monitor, b>=0

The aim: z = x

b+(1-b)y^G = x

b+(1-b)f(x)^G = x

f(x) = [(x-b)/(1-b)]^(-G)

A power function cannot have a negative base.

1) Replace f(x)=0 for x less than b
This will cause clipping for dark values,
the rest is correct.

2) Replace f(x)=x^(-G)
The blackpoint is ignored (b=0).
Then we have no clipping and
z=b+(1-b)(x^(-G))^G = b+(1-b)x
This is an inherent BPC, without using the
blackpoint.

My conclusion: CMS doesn't need any knowledge
about the monitor blackpoint. The results are
as expected and fairly good if (b) is small.
And it explains, why my ProfileMaker monitor
profiles don't contain the tag.

Best regards --Gernot Hoffmann

Sorry,

idiotic mistake. Correction:

y(x) = f(x)
This is a so far unknown function

z(y) = b+(1-b)y^G
TRC for the calibrated (!) monitor, b>=0

The aim: z = x

b+(1-b)y^G = x

b+(1-b)f(x)^G = x

f(x) = [(x-b)/(1-b)]^(1/G)

A power function cannot have a negative base.

1) Replace f(x)=0 for x less than b
This will cause clipping for dark values,
the rest is correct.

2) Replace f(x)=x^(1/G)
The blackpoint is ignored (b=0).
Then we have no clipping and
z=b+(1-b)(x^(1/G))^G = b+(1-b)x
This is an inherent BPC, without using the
blackpoint.

G.H.

Gernot, Where did you get that equation b+(1-b)y^G? Do actual monitors behave exactly as that equation would indicate and also does your equations relate to the 1D LUT tables used by video cards to calibrate the monitor to the workspace gamma? Your equations are new to me have to admit I am not knowledgable enough. However have you read <http://tinyurl.com/j52xy> of Larry Tseng especially with how PhS predict its source black and destination black from the profiles. I still have to complete that pdf file as I have never read it before and don't know if the BPC there does introduce more black clippings for softproofing default to using Relative Intent with BPC?

We are getting more and more deeper into small details and really takes me time to digest but very interesting.

Kwanyue,

my model for calibrated (!) CRT monitors is based on
experiments, after reading other sources:
<http://www.fho-emden.de/~hoffmann/measgamma10022004.pdf>

Please be aware, that the dark level in each channel
can hardly be measured accurately with our common
instruments. The calculated dark levels and exponents
are somewhat uncertain, especially in the blue channel.

For the arguments in my previous letter, my model doesn't
depend on an accurate power function with G=2.2.
The arguments would be the same, if each TRC were given by
an offset plus an invertible table function.

I know the quoted doc. It explains nicely how BPC works
for printer profiles.
In the moment I'm preferring for monitors my new explanation.

I'm anyway not trying to explain what's happening in PhS.

Your original question (first post):

'Why not just Relative Intent alone without BPC?
I was talking about softproofing colors of ICC
profiles on monitor. Why the NEED to map black
point of the ICC to absolute zero RGB black of
the monitor ICC? Why not just accurately simulate
the black point of the ICC profile from the monitor?'

The anwer is IMO: sending the working space data (which have
XYZ blackpoint zero) ONLY with gamma correction to the monitor
results automatically in BPC. We can see everything 'on top
of the actual black level'. That's the same as outputting
data for a printer 'on top of the printer's black level'.
But for the monitor one doesn't need to know the BP - that's
the difference.

If you accept my model - would you (and Larry) agree to my
conclusions ?

Best regards --Gernot Hoffmann

Gernot, this is your calibration model. But how about the calibration model of the monitor profiler software which may not calibrate according to your monitor TRC model? Hmmm... am I asking my question correctly?

Kwanyue,

that's simple - the system identification (profiling)
by least square methods or any other optimization
depends on the underlying model.

I'm feeling happy, if the measured data points are
interpreted by my program as Gamma=2.17, whereas the
professional software may say Gamma=2.21.
Now use different software packages. I'm sure that the
results will be different.

IMO: doesn't matter, as long as the white point (D65)
is almost the same for varying (higher)luminances.

Occasionally we should go back to the roots:
We are editing images in order to prepare them for
printing. I'm assuming that's your task as well.
Where is the bottleneck in the whole workflow ?
Probably not the issue which we are discussing here.
Black level 0.5 or 0.3 or 0.2 cd/m2 ...
Really important ? I dont' think so. The bottleneck
is IMO the reproduction of neutral grays with suffi-
cient level resolution in the shadows by any printer.

Best regards --Gernot Hoffmann

Gernot,

Regarding your:

2) Replace f(x)=x^(1/G)
The blackpoint is ignored (b=0).
Then we have no clipping and
z=b+(1-b)(x^(1/G))^G = b+(1-b)x
This is an inherent BPC, without using the
blackpoint.

If y(x) = f(x)= x^(1/G)
then x not equal z
right?

However is not traditional convention is to ignore the b and so that y(x) = f(x)= x^(1/G) and also x = z?