Re: [Lcms-user] Preview of black preserving feature on CVS

[Marti <ma...@li...>]

Hi Gerhard,

Please bear in mind this is still a work in progress. There is almost nothing 
on that on open source, so I am still trying to figure out what would be the 
"good" way to do things. Black-preserving transforms are not well assessed 
by dE metrics. There are some situations where dE is not such important, 
and this is one of these.

> The chromaticity of the src and dst K-only axis may differ a couple of 
> dE_ab units. If [0,0,0,K] is mapped to [0,0,0,K'], but e.g. 
> [0.001,0,0,K] is mapped to [C',M',Y',K'], such that [0.001,0,0,K] and 
> [C',M',Y',K'] match colorimetrically, doesn't this produce a 
> discontinuity immediately near the K-only axis? Shouldn't there be done 
> a smooth blending? (sure, the limited CLUT resolution will smooth the 
> transition, but is this sufficent?).

Note that a couple of dE in the black ink means a huge tint. 
That's very evident because the eye is far more sensitive near neutral 
axes than in any other region. If the original separation is correct, 
this effect greatly improves overall appearance. 

For sure you have seen prints of gray images with a huge 
color cast towards magenta or green. That's very unpleasant. 

Mainly, this happens because metamerism is not taken into 
account when doing the profile. Black ink chromaticity  changes on 
different illuminants. For example, a profile is done measuring black 
ink under D50, then, under D50 this black ink have tendency to 
magenta. Ok, the profile captures such chroma and, when
reproducing colorimetrically, replaces the destination black with
CMY reproducing this magenta. Now, If you take the original K ink
and examine it under sunlight, it is not magenta anymore! But the 
reproduction using CMY keeps going magenta. Result: a nasty color 
cast . 

And this is just one of the reasons why keeping black ink is so important. 
Maybe there is a slight discontinuity, as big as the chromaticity of
blacks differ, but it is so small that the smoothing induced by the 
CLUT is enough to compensate. And this is almost nothing when 
compared with the huge cast on grays a CMYK->Lab->CMYK 
may create.


> If [0,0,0,K] is mapped to [0,0,0,K'] in the destination color space, 
> shouldn't also [C,M,Y,0] be mapped to [C',M',Y',0], such that [0,0,0,K'] 
> and [C',M',Y',0] (in the destination color space) have the same color as 
> well?

That's almost the same question. CMY=gray is not maintained anymore, but
a colorimetric match. If you place, side by side, two gray patches; one  done
by using equal amounts of CMY and other done exclusively by K ink, then
after a black preserving transform you should end with a patch with same 
chromaticity as equal amounts of CMY in the source profile, and a patch 
created by using K ink only. To what extent the CMY is still gray... well, it 
depends. If the origin profile said a and b were close to zero, you are going to 
obtain a,b close to zero on output but only under D50. In real world, most times 
illuminants are  not D50, and you end with some sort of tint. That doesn't 
happen on the K only patch.

> Isn't there a risk, that the Newton method gets stuck in a local minimum?

Yes, and it happens. I am still not fully satisfied by the algorithm. It works to 
some extent, but misses ink limiting (TAC) preservation. I have to figure out
how to get the TAC from the output profile and how to avoid local minimums
as well.

Best regards,
Marti.


----- Original Message ----- 
From: "Gerhard Fuernkranz" <nos...@gm...>
To: <lcm...@li...>
Sent: Tuesday, May 24, 2005 1:25 AM
Subject: Re: [Lcms-user] Preview of black preserving feature on CVS


> Marti schrieb:
> 
> Hi Marti,
> a few comments/questions below.
> 
>> The algorithm is a little bit more complex. Here is a coarse outline:
>>
>> 1) A K tone curve from K-> K' is computed. That is, which amount of K'
>> on the output space gives same L* as which amount of K in input. That 
>> is done by measuring ramps of (0,0,0, K) using the proof direction,
>> on both input and output profile. Then joining the obtained curves.
>>
>> 2) The tone curve is used on (0,0,0,K) -> (0,0,0,K') so pure black is 
>> always preserved as pure black.
> 
> The chromaticity of the src and dst K-only axis may differ a couple of 
> dE_ab units. If [0,0,0,K] is mapped to [0,0,0,K'], but e.g. 
> [0.001,0,0,K] is mapped to [C',M',Y',K'], such that [0.001,0,0,K] and 
> [C',M',Y',K'] match colorimetrically, doesn't this produce a 
> discontinuity immediately near the K-only axis? Shouldn't there be done 
> a smooth blending? (sure, the limited CLUT resolution will smooth the 
> transition, but is this sufficent?).
> 
> Another related issue, where IMO the desired behaviour is not so clear, 
> and different people may have a different opinion:
> 
> Example:
> Assume that [0,0,0,K] and [C,M,Y,0] have the same color in the source 
> color space.
> 
> If [0,0,0,K] is mapped to [0,0,0,K'] in the destination color space, 
> shouldn't also [C,M,Y,0] be mapped to [C',M',Y',0], such that [0,0,0,K'] 
> and [C',M',Y',0] (in the destination color space) have the same color as 
> well?
> 
> Or in other words:
> If two (different) CMYK values have the same color in the source color 
> space, wouldn't one expect, that the colors of the transformed CMYK 
> values also match in the destination color space?
> 
>> 2) AToBXX tags on output profile are reversed and a special BKToA is 
>> computed. This is a new LUT has a extended PCS using Lab + K.
>>
>> 3) Then a devicelink is built by using the original AtoB plus this new 
>> BKToA.  Original K is passed to the extended PCS across tone curve.
>>
>> The method for inversion is a modified Newton-Raphson extended to 3 
>> dimensions. The original CMY + K across tone curve is used as a seed for
>> the search. The algorith iterates across Jacobian as far as the error 
>> decreases and the system is convergent.
> 
> Isn't there a risk, that the Newton method gets stuck in a local minimum?
> 
>> This have some nice side effects:
>>
>> - If the color cannot be matched at all, the original separation is 
>> keept as much as possible.
> 
> I've seen, you have again removed the MaximumError threshold and the 
> different handling below and beyond the threshold. That's IMO good, 
> since it also did introduce a discontinuity at the threshold.
> 
> Regards,
> Gerhard
> 
> 
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by Oracle Space Sweepstakes
> Want to be the first software developer in space?
> Enter now for the Oracle Space Sweepstakes!
> http://ads.osdn.com/?ad_id=7412&alloc_id=16344&op=click
> _______________________________________________
> Lcms-user mailing list
> Lcm...@li...
> https://lists.sourceforge.net/lists/listinfo/lcms-user
> 
> 
> 
> -- 
> No virus found in this incoming message.
> Checked by AVG Anti-Virus.
> Version: 7.0.322 / Virus Database: 266.11.16 - Release Date: 24/05/2005
>


-- 
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.322 / Virus Database: 266.11.16 - Release Date: 24/05/2005