Regarding fitting of Shaper / LAB / XYZ input profiles (question to Graeme)

[Derin Korman]

Hello Graeme,
First of all, thank you so much for all the work, I use your software all
the time and have just started using it for some professional work, so I
will make sure to donate sometime soon. If you'd like, I can help with
reformatting the website a bit for increased legibility also, without
touching the content.

Anyhow, to the question.
For input profiles such as scanners, could you expand on the processes of
the different profile types, as to how they go from the input/reference
data to the cLUT / curves? I am assuming that the shaper is higher
polynomials smoothed by lower order ones to create io curves for each
channel, but I don't understand if XYZ / LAB employ some sort of smoothing
to fit the cLUT itself, or whether the CMM that uses the profile for
conversion uses its own smoothing based on the look up table data.

Thank you very much!
*ps it could be helpful to have these on the website too maybe?

*~d*

[Graeme Gill]

Post by Derin Korman
For input profiles such as scanners, could you expand on the processes of
the different profile types, as to how they go from the input/reference
data to the cLUT / curves?

Hi,
the ICC standard itself has an introduction that outlines
the various steps in different ICC profile configurations (see section 0.4).
Note that this includes the ICC V4 combinations that Argyll doesn't currently
support.

Post by Derin Korman
I am assuming that the shaper is higher
polynomials smoothed by lower order ones to create io curves for each
channel,

That's the type of approach I've used in creating ArgyllCMS shaper
curves, but it's not a limitation of the profile format. Reducing
the curves to a moderate number of parameters makes them able to
be set by a fitting process. I don't use polynomial basis functions
though, but something more appropriate to a transfer curve that
is constrained to be monotonic.

Post by Derin Korman
but I don't understand if XYZ / LAB employ some sort of smoothing
to fit the cLUT itself, or whether the CMM that uses the profile for
conversion uses its own smoothing based on the look up table data.

Typically the CMM doesn't modify the cLUT, but uses it as-is.

In doing a scattered data fit it is necessary to use some sort
of regularization to guaranteed continuity in areas that are
otherwise not constrained by measured data point. A smoothness
constraint provides this, as well as a mechanism to reduce the
effect of uncertainties in the measurement values.

Post by Derin Korman
*ps it could be helpful to have these on the website too maybe?

Sorry, I'm not clear on how to label such information, and I'm
not sure how broad interest is in such details. Typically such
information is in the source code comments.

Graeme Gill.