Digital BW, The Print

In a message dated 2/27/06 12:57:45 AM, odesmais@... writes:


> Sorry if I was unclear. QTR is quite well documented, so one knows 
> perceptual is to be used.
> 
> What I can't figure out due to limited understanding is WHY perceptual 
> instead of relative is being implemented by Roy.
> 
> It's about pure personal understanding...
> 
> 

I would guess because its the most common default. So it would be selected, 
not because its actually related to what the software's doing, intentwise; but 
because its the default intent that is most commonly autoselected, and 
therefore the best choice for single intent functions.

C. David Tobie
Product Technology Manager
ColorVision Business Unit
Datacolor Inc.
CDTobie@...
www.colorvision.com


[Non-text portions of this message have been removed]

David,

The way I understand relative is that it somehow does a clipping of 
out-gamut values without re-evaluating in-gamut values oppositely to 
perceptual that compresses them all.

Extrapolating this to L*, while it still runs from media white to 
media black (with BPC) in both case, L* would be re-scaled in 
perceptive and not in relative ?

I know it doesn't add anything to anyone's workflow, I just can help 
thinking about it without finding out and get obsessed...

As said, I'm new to this so thank for being patient...

Olivier

 
> In a message dated 2/27/06 12:57:45 AM, odesmais@... writes:
> 
> 
> > Sorry if I was unclear. QTR is quite well documented, so one 
knows 
> > perceptual is to be used.
> > 
> > What I can't figure out due to limited understanding is WHY 
perceptual 
> > instead of relative is being implemented by Roy.
> > 
> > It's about pure personal understanding...
> > 
> > 
> 
> I would guess because its the most common default. So it would be 
selected, 
> not because its actually related to what the software's doing, 
intentwise; but 
> because its the default intent that is most commonly autoselected, 
and 

> therefore the best choice for single intent functions.
> 
> C. David Tobie
> Product Technology Manager
> ColorVision Business Unit
> Datacolor Inc.
> CDTobie@...
> www.colorvision.com
> 
> 
> [Non-text portions of this message have been removed]
>

More accurately, Roy and people like myself who were prodding him along
weren't able to get particularly clear answers as to how the various intents
- well relcol and perceptual in particular - OUGHT to be applied in the B&W
case.  So Roy had to piecemeal together information from scattered sources.
For example, some may recall the initial difficulty Roy had with clipping
highlights.  We latter "discovered" all input data had to be scaled for
media relativity, ie media white point, a transform in CIE-XYZ.  Roy found
an article by Adobe which described their black point compensation at a very
high level.  Again another scaling in XYZ space.  BPC plugs a gap in the ICC
spec.  So then you've got (media) relative colormetric with BPC.  The
perceptual intent allows a free hand to the profile engineer.  However,
whenever we attempted to ask the question:  "how would you, given the
freedom of the perceptual intent, apply such freedom to a simple B&W
scenario?" we didn't get many, if any, answers.  Roy will correct me if I am
wrong, but he has essentially applied relcol+bpc and slotted it in the
perceptual tags.  It was slotted in the Perceptual tags because the
intention from the beginning was to develop an appropriate perceptual
transform.  In a later iteration that he mentioned in an earlier post, he
filled out the other intent tags with the same transform because of some
uncertainty as to how PS used these others tags when establishing soft
proofs.  Perceptual intent formulae tend to be closely guarded secrets:  the
secret sauce that distinguishes one profiling package from another.
Nonetheless, since none attend to the needs of the B&W photographer (not
using a colour workflow) we would greatly appreciate the expertise that
perhaps someone like David could provide. I think it's fair to say that we
are very happy with the transform currently embedded in a QTR ICC profile,
particularly given that it's affect can be soft proofed, but would like to
find out more about whether or not alternative transforms may be useful or
more "scientifically correct".

> From: <CDTobie@...>
> Reply-To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Date: Mon, 27 Feb 2006 10:39:19 EST
> To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Subject: Re: [Digital BW] Re: Questions about Using Color Settings and
> Workspace Profiles for Grayscale Printing
> 
> 
> In a message dated 2/27/06 12:57:45 AM, odesmais@... writes:
> 
> 
>> Sorry if I was unclear. QTR is quite well documented, so one knows
>> perceptual is to be used.
>> 
>> What I can't figure out due to limited understanding is WHY perceptual
>> instead of relative is being implemented by Roy.
>> 
>> It's about pure personal understanding...
>> 
>> 
> 
> I would guess because its the most common default. So it would be selected,
> not because its actually related to what the software's doing, intentwise; but
> because its the default intent that is most commonly autoselected, and
> therefore the best choice for single intent functions.
> 
> C. David Tobie
> Product Technology Manager
> ColorVision Business Unit
> Datacolor Inc.
> CDTobie@colorvision.com
> www.colorvision.com
> 
> 
> [Non-text portions of this message have been removed]
> 
> 
> 
> Please visit the Group Homepage to check the Files, and other resources as
> they are often being updated.
> 
> http://groups.yahoo.com/group/DigitalBlackandWhiteThePrint
> 
> If you wish to receive no emails or just a daily digest, or you wish to
> unsubscribe, please edit your Membership preferences by visiting this same
> page.
> 
> Please follow these basic guidelines:
> - As threads develop, trim off excess portions of earlier messages to keep
> them short.
> - Good manners are required at all time. No personal attacks or flames.
> Hostile, aggressive or argumentative users may be removed from the membership
> without notice.
> - Keep your posts and threads related to the group topic of digital B&W
> printing. Users who persistently make off-topic posts may be removed from the
> membership.
> - By posting on this forum you agree to abide by the group rules and
> guidelines, and to abide by the actions and decisions of the group Owner and
> Moderators. See „Group Topic, Rules and Guidelines‰ in the Files section:
> http://groups.yahoo.com/group/DigitalBlackandWhiteThePrint/files/
> 
> BY PARTICIPATING IN AND/OR POSTING MESSAGES TO THE DIGITAL BW, THE PRINT
> YAHOO! GROUP YOU EXPRESSLY UNDERSTAND AND AGREE THAT THE „OWNER‰ AND
> „MODERATORS‰ OF DIGITAL BW, THE PRINT YAHOO GROUP SHALL NOT BE LIABLE TO YOU
> FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL OR EXEMPLARY
> DAMAGES, INCLUDING BUT NOT LIMITED TO, DAMAGES FOR LOSS OF PROFITS, GOODWILL,
> USE, DATA OR OTHER INTANGIBLE LOSSES (EVEN IF THE  „OWNER‰ AND „MODERATORS‰ OF
> DIGITAL BW, THE PRINT YAHOO GROUP HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH
> DAMAGES), RESULTING FROM: (i) THE USE OR THE INABILITY TO USE THE DIGITAL BW,
> THE PRINT YAHOO GROUP; (ii) UNAUTHORIZED ACCESS TO OR ALTERATION OF YOUR
> TRANSMISSIONS OR DATA; (iii) STATEMENTS OR CONDUCT OF ANY THIRD PARTY ON THE
> DIGITAL BW, THE PRINT YAHOO GROUP; OR (iv) ANY OTHER MATTER RELATING TO THE
> DIGITAL BW, THE PRINT YAHOO GROUP.
>  
> Yahoo! Groups Links
> 
> 
> 
>  
> 
> 
>

From Olivier:

"David,

The way I understand relative is that it somehow does a clipping of
out-gamut values without re-evaluating in-gamut values oppositely to
perceptual that compresses them all.

Extrapolating this to L*, while it still runs from media white to
media black (with BPC) in both case, L* would be re-scaled in
perceptive and not in relative ?

I know it doesn't add anything to anyone's workflow, I just can help
thinking about it without finding out and get obsessed...

As said, I'm new to this so thank for being patient...

Olivier"

This is why BPC fills a gap in the ICC spec.  The greyscale axis is scaled
for media white and then for ink black.  Without BPC, blacks would get
clipped. Out of gamut colours that do not sit on the greyscale axis are then
amended to the closest possible value.  So there is a "shrinking of the
gamut" as far as greyscale is concerned - the whole scale is compressed to
fit.  The question for the perceptual intent is whether Adobe BPC and simple
media white point scaling are the most visually optimal methods of shrinking
the gamut to fit?  (There then exists another much more complex question as
to whether this would remain constant for "greyscales" of varying hue -
warm, cool, sepia etc.)

> From: Steve Kale <stevekale@...>
> Reply-To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Date: Mon, 27 Feb 2006 16:09:03 +0000
> To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Conversation: [Digital BW] QTR Create ICC tranforms:  was Questions about
> Using Color Settings and Workspace Profiles for Grayscale Printing
> Subject: Re: [Digital BW] QTR Create ICC tranforms:  was Questions about Using
> Color Settings and Workspace Profiles for Grayscale Printing
> 
> More accurately, Roy and people like myself who were prodding him along
> weren't able to get particularly clear answers as to how the various intents
> - well relcol and perceptual in particular - OUGHT to be applied in the B&W
> case.  So Roy had to piecemeal together information from scattered sources.
> For example, some may recall the initial difficulty Roy had with clipping
> highlights.  We latter "discovered" all input data had to be scaled for
> media relativity, ie media white point, a transform in CIE-XYZ.  Roy found
> an article by Adobe which described their black point compensation at a very
> high level.  Again another scaling in XYZ space.  BPC plugs a gap in the ICC
> spec.  So then you've got (media) relative colormetric with BPC.  The
> perceptual intent allows a free hand to the profile engineer.  However,
> whenever we attempted to ask the question:  "how would you, given the
> freedom of the perceptual intent, apply such freedom to a simple B&W
> scenario?" we didn't get many, if any, answers.  Roy will correct me if I am
> wrong, but he has essentially applied relcol+bpc and slotted it in the
> perceptual tags.  It was slotted in the Perceptual tags because the
> intention from the beginning was to develop an appropriate perceptual
> transform.  In a later iteration that he mentioned in an earlier post, he
> filled out the other intent tags with the same transform because of some
> uncertainty as to how PS used these others tags when establishing soft
> proofs.  Perceptual intent formulae tend to be closely guarded secrets:  the
> secret sauce that distinguishes one profiling package from another.
> Nonetheless, since none attend to the needs of the B&W photographer (not
> using a colour workflow) we would greatly appreciate the expertise that
> perhaps someone like David could provide. I think it's fair to say that we
> are very happy with the transform currently embedded in a QTR ICC profile,
> particularly given that it's affect can be soft proofed, but would like to
> find out more about whether or not alternative transforms may be useful or
> more "scientifically correct".

The question for the perceptual intent is whether Adobe BPC and simple
> media white point scaling are the most visually optimal methods of 
shrinking
> the gamut to fit?  (There then exists another much more complex 
question as
> to whether this would remain constant for "greyscales" of varying 
hue -
> warm, cool, sepia etc.)
> 

Well, Steve the question as I can best formulate it would be : would 
relcol simply scale L* from media white to BPCted black in a 
linearmanner (the way you linearise the printer output e.g. evenly 
spaced L* intervals) while the perceptual one would include a gamma-
like formula to the restricted L* (L* being scaled only from media 
white to BPCted black instead of 0-100)thus preserving human 
perception and contrast.

Sorry I can't think in XYZ, and all the above probably sounds very 
naive...


Olivier

Not at all.  You're going through a process of thinking that I have gone
through over the last year or so.  I don't have the answers and your
questions probe exactly the gaps in my understanding but here's a start.
The CIE XYZ primary system attempts to model our human sight and XYZ_Y is
the "primary" which seeks to measure luminance IN THE WAY THE EYE SEES IT.
To quote Bruce Fraser in Real World Color Management 2nd Edition pg 41:

"...the primary Y doubles as the average luminance function of the cones -
so a color's Y value is also its luminance"

Y is a rough approximation of our logarithmic response to luminance.  The
uniform colour spaces of LAB and LUV were attempts to create colour spaces
that are perceptually uniform.  This doesn't mean they model the eye any
better but rather that "distances between points in the space predict how
different the two colours will appear to a human observer" (pg 42)  L* is
approximately the cube root of the luminance value Y.

I'm not a mathematician but I suspect that the scaling could be performed in
L*, since there is of course a relationship between L* and XYZ_Y, but if a
linear scaling of luminance in XYZ_Y is required then it's unlikely to be a
linear scaling in L*.  Roy can likely fill in the gaps as to why it's linear
in XYZ (and hence not linear in L*) and I think this is the missing piece in
your thinking (and mine).

You need to separate your thinking of linearising a printer's response and
optimal tonal mapping and compression from the image file to the dynamic
range of ink on paper.  Having very linear L* output from a printer is NOT a
satisfactory end-game because on its own it does not generate visually
pleasing renditions of images, particularly when we can't get decent blacks
(such as on matte paper).  Prints are typically "flat" and generally
"lighter than on screen". This is the very issue which drove Roy to think
about the whole topic of applying ICC methodologies to B&W printing.
Getting a B&W printer to be linear is good because it means that the printer
has a predictable response across the scale of possible values.  But we
still need a suitable formula or mechanism for mapping from our image file
space to this linear L* print space. The advantage of using an ICC profile
is that the CMM "appropriately" does the transform from any document space
to the print space.  We are no longer dependent on individual transforms for
each and every combination of workspace and print space.

As I write this I realise I do not have a full and clear understanding of
mathematics of the scaling that is done for media relative colorimetry and
Adobe black point compensation - specifically why a linear scaling in XYZ is
applicable.  I hope Roy can fill in some of the gaps.

Cheers

Steve

> From: Olivier <odesmais@...>
> Reply-To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Date: Mon, 27 Feb 2006 16:56:24 -0000
> To: <DigitalBlackandWhiteThePrint@yahoogroups.com>
> Subject: Re: [Digital BW] QTR Create ICC transforms:  was Questions about
> Using Color Settings and Workspace Profiles for Grayscale Printing
> 
> The question for the perceptual intent is whether Adobe BPC and simple
>> media white point scaling are the most visually optimal methods of
> shrinking
>> the gamut to fit?  (There then exists another much more complex
> question as
>> to whether this would remain constant for "greyscales" of varying
> hue -
>> warm, cool, sepia etc.)
>> 
> 
> Well, Steve the question as I can best formulate it would be : would
> relcol simply scale L* from media white to BPCted black in a
> linearmanner (the way you linearise the printer output e.g. evenly
> spaced L* intervals) while the perceptual one would include a gamma-
> like formula to the restricted L* (L* being scaled only from media
> white to BPCted black instead of 0-100)thus preserving human
> perception and contrast.
> 
> Sorry I can't think in XYZ, and all the above probably sounds very
> naive...
> 
> 
> Olivier

Steve,

Thanks for the lengthy explanation.
I assume one has to go through the full reading of ICC specs. Though I 
strongly doubt if I do so, I'll ever come any close to the 
understanding not being a mathematician either (reason why I was hoping 
to have someone explain me the process in an everyday wording).

It's frustrating to click on the mouse without understanding what's 
on...!

Olivier

In a message dated 2/27/06 11:07:45 AM, odesmais@... writes:


> The way I understand relative is that it somehow does a clipping of
> out-gamut values without re-evaluating in-gamut values oppositely to
> perceptual that compresses them all.
> 

That would be a typical description of the difference between those intents. 
But that refers to out of gamut colors, not blacks and whites, which are dealt 
with differently. And here we are speaking of a specialty black and white 
product, which can choose to use the differing intents any way it sees fit, if it 
chooses to offer more than one option.

C. David Tobie
Product Technology Manager
ColorVision Business Unit
Datacolor Inc.
CDTobie@...
www.colorvision.com


[Non-text portions of this message have been removed]

Hello CD.
Can you expand on the, dealt with differently, point? What about
nonspecialty use?

Regards
Duane

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, CDTobie@... wrote:

> But that refers to out of gamut colors, not blacks and whites, which
are dealt 
> with differently. And here we are speaking of a specialty black and
white 
> product, which can choose to use the differing intents any way it
sees fit, if it 

> chooses to offer more than one option.
> 
> C. David Tobie
> Product Technology Manager
> ColorVision Business Unit
> Datacolor Inc.
> CDTobie@...
> www.colorvision.com
> 
> 
> [Non-text portions of this message have been removed]
>

That precisely where there's confusion for me : to my understanding perceptual remaps ALL values regardless of them being in/out-gamut oppositely to relative.

If right, greyscale should also have a tranform which is different in both intents. But... as you said greyscale is dealt differently : I'd be grateful if you could further elaborate on this.

Olivier

> > 
> 
> That would be a typical description of the difference between those intents. 
> But that refers to out of gamut colors, not blacks and whites, which are dealt 
> with differently. And here we are speaking of a specialty black and white 
> product, which can choose to use the differing intents any way it sees fit, if it 
> chooses to offer more than one option.
> 
> C. David Tobie
> Product Technology Manager
> ColorVision Business Unit
> Datacolor Inc.
> CDTobie@...
> www.colorvision.com
> 
> 
> [Non-text portions of this message have been removed]
>

In a message dated 2/27/06 2:52:09 PM, dlruckus@... writes:


> Hello CD.
> Can you expand on the, dealt with differently, point? What about
> nonspecialty use?
> 
> Regards
> Duane
> 
> --- In DigitalBlackandWhiteThePrint@yahoogroups.com, CDTobie@... wrote:
> 
> > But that refers to out of gamut colors, not blacks and whites, which
> are dealt
> > with differently. And here we are speaking of a specialty black and
> white
> > product,
> 

Intents deal with out of gamut colors. Paper white and ink black compensation 
work with "out of gamut" blacks and whites instead. There are no out of gamut 
colors (or any colors to speak of) in gray systems. But you still need to 
stretch the profile to print from paper white to ink black, not clip the ends as 
it would without these adjustments. If you look at the Ref Black and Ref White 
functions in PrintFIX PRO you will see that these allow adjustment for 
differing degrees of preview adjustment to paper white (both warmth, and brightness) 
and ink black (tint matters less here, its mostly darkness). These functions 
are to fine tune the white and black (and everything in between, as effected 
by white and black) of your softproof preview to improve your screen to print 
match, without actually changing how the profile prints. Same idea, but for 
color work, or for black and white printed with color profiles.

C. David Tobie
Product Technology Manager
ColorVision Business Unit
Datacolor Inc.
CDTobie@...
www.colorvision.com


[Non-text portions of this message have been removed]

In a message dated 2/27/06 4:14:27 PM, odesmais@... writes:


> That precisely where there's confusion for me : to my understanding 
> perceptual remaps ALL values regardless of them being in/out-gamut oppositely to 
> relative.
> 

Thats possible, but rather blind and reckless, so its not the way all 
developers choose to go... since it happens with no knowledge of whether compression 
is appropriate or desirable, for the device in question.

C. David Tobie
Product Technology Manager
ColorVision Business Unit
Datacolor Inc.
CDTobie@...
www.colorvision.com



[Non-text portions of this message have been removed]

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, "Olivier" <odesmais@...> 
wrote:
>
> The question for the perceptual intent is whether Adobe BPC and simple
> > media white point scaling are the most visually optimal methods of 
> shrinking
> > the gamut to fit?  (There then exists another much more complex 
> question as
> > to whether this would remain constant for "greyscales" of varying 
> hue -
> > warm, cool, sepia etc.)
> > 
> 
> Well, Steve the question as I can best formulate it would be : would 
> relcol simply scale L* from media white to BPCted black in a 
> linearmanner (the way you linearise the printer output e.g. evenly 
> spaced L* intervals) while the perceptual one would include a gamma-
> like formula to the restricted L* (L* being scaled only from media 
> white to BPCted black instead of 0-100)thus preserving human 
> perception and contrast.
> 
> Sorry I can't think in XYZ, and all the above probably sounds very 
> naive...
> 
> 
> Olivier
>

Olivier,

Both Steve and C. David have said a lot, maybe I can add a little.

XYZ is really pretty simple as far as B&W is concerned.  All the B&W info is
in the Y component.  It's called Luminance and it's basically a measure of the
total amount of light -- there's no "color" in it.   The measure is the real "physics"
measure of light.  I.e. you can think of it as "how many photons" or "how much
energy is reflected".   It's normalized to 0 to 1, pure black would be 0 i.e. nothing
is reflected, pure white would be 1 i.e. all the incident light would be reflected.
In real life you never get to either endpoint -- there is no pure black nor pure white.

Scaling calculations using Y (XYZ) is used thoughout all the ICC documents and
when you think of it, it makes a lot of sense.  Think of a print illuminated with a
specific light.  There's a number of photons reflected off various parts of the print
based on the densities.  If you reduce the illuminating light by half (.5 x photons)
all the reflections will also be reduced by half.  This is basically linear scaling
in Y units.  Print illumination ought to be invariant so linear scaling should be
invariant too.   This basically goes with the media-white-relative definitions in ICC
profiles i.e. media white is Y=1. 

Back to profiles, with B&W the given (I think) is that black has to map to dMax or
black ink and white to map to dMin or paper white.  Anything different will be worse.
Because of the above, scaling Y to paper white is almost a given.  What to do with
black however is not nearly as obvious.  In fact the ICC specs don't address it.
Adobe came up with a concept they have called Black Point Compensation.
Mathematically it turns out to be just the same linear Y scaling at the black end
that was already used at the white end.  Mathematics aside the issue is: does this
give good results?

When I first started QTR, "linearizing" from dMax to dMin was an obvious need.
But what is linear is not at all clear.   There's a function called L* which takes the
Y luminance above and makes it "perceptually" linear.  I.e L values match human
perception.   So for QTR, linear L-values are how all the linearization is done.
This might seem the best since we take the grayscale values and spread them out
with the best perceptual separation.  However in practice there are two issues.
First, on the file side there are grayscale embedded profiles that are all different
so it's impossible to match them all.  Second, even if you make a even L-value
embedded profile (that's the Gray Lab profile) when you look at the screen
versus the prints it just doesn't match that well.  This is mostly an issue with 
matte papers where dMax is not very dense.  The prints always looked lighter 
than the screen.

This is where trying to figure out mapping all the values between dMax and dMin
starts.   Implementing the BPC that Adobe defined turns out to match screen and
print a whole lot better than before.  And putting it all in an ICC profile solves
the issue (1) above.  It's probably not perfect since the blacks on the screen and 
paper are so different.

So this is where all the mapping originates.  I think the name "perceptual" seems
to fit the issue.  Because of color, ICC profiles have 3 intents but I don't think
there's any real difference as far as B&W, but coincidentally one of the intents 
has the name perceptual.  The intents sort of sneak into the issue in soft-proofing
because of the conversions into the screen profile.

Olivier, back to this question:
> Well, Steve the question as I can best formulate it would be : would 
> relcol simply scale L* from media white to BPCted black in a 
> linearmanner (the way you linearise the printer output e.g. evenly 
> spaced L* intervals) 
It would seem to be a reasonable idea, but that's basically how the QTR linearization
works and the screen to print match doesn't work as well.

Roy