Re: [Lcms-user] Linearization

[Tamas L. <lit...@gm...>]

Thank you for your fast reply!

It took me a few days to find the bug in the "reverse interpolation" part,
and you are right, I was lost in details.
Probably I can fine-tune the interpolation somehow, but the printed results
- especially the "BullsEye" gradients are very beautiful now.

Thanks again!
Tamas


On Thu, 15 Jun 2023 at 08:41, <mar...@li...> wrote:

> Hello,
>
>
>
> When doing this kind of things, it is important to keep in mind the real
> goals of the process because sometimes is very easy to get lost in details.
>
>
>
> “Linearization” is a term that I’ve seen used to a multitude of scenarios.
>
>
>
>
> A valid approach when profiling a printer is to do it on two steps. The
> first step would be to bring the particular printer to a generic well-known
> state. For example, by using C, M Y and K tone curves, one per output
> channel. Then profile this generic printer.  The first step is known as
> “linearization”. The obvious advantage of this approach is you have not to
> reprofile your printer often but just recompute the curves, which is by far
> cheaper and faster.
>
>
>
> Another situation when the term linearization is used is when you want a
> profile to “match” the characteristics of a given metric. Examples of this
> are to create an RGB device space that behaves linearly on Luma.  Some of
> the linearization goal metrics are intuitive, others are not. For example,
> when linearizing CMYK, you can use L* on all channels but on Yellow.
> Reason is yellow contributes poorly to L*, and is better to use b* in this
> case. Other metrics can be optical density or its derivatives.
>
>
>
> So, assuming you want a tone curve to make a gray device space to behave
> linearly with L*, you could print a scale of 10-20 patches and then measure
> it. With the obtained values perform reverse interpolation. For each L* you
> will obtain which device gray value to print. With those values build a
> tone curve and you got the linearization. As a test, your tone curve should
> match the printed values.
>
>
>
> When doing multidimensional, things go more complex though.
>
>
>
> Hope that helps
>
> Regards
>
>
>
> Marti Maria
>
> The LittleCMS Project
>
> https://www.littlecms.com
>
>
>
>
>
>
>
>
>
>
>
>
>
> *From:* Tamas Littmann <lit...@gm...>
> *Sent:* Wednesday, June 14, 2023 1:39 PM
> *To:* lcm...@li...
> *Subject:* [Lcms-user] Linearization
>
>
>
> Hello,
>
>
>
> I have used LCMS for image color space transforms for lots of years, but I
> am relatively new in creating my own profiles.
>
> Using C# and .NET, I can do most of the creating process successfully.
>
> However, it seems that I cannot move on from some of the easiest steps (I
> think I understand the process, but the results are not satisfactory).
>
>
>
> Here, I try to minimize the problem, and use 1D transforms (linearization
> curve) and Grayscale space (only 1 color).
>
>
>
> In this case, how can I linearize a black&white grayscale image/printer?
>
>
>
> Here is what I tried to do:
>
> 1. Print an equal step  grayscale ramp (in grayscale mode, using 8-bit 0 -
> 255 values. E.g: for six steps it's 0/51/102/153/204/255). 52 steps seems
> to be a good starting point (steps are 0 - 5 - 10 - 15 - 20 - etc..)
>
>
>
> 2. I have an X-Rite i1 / Datacolor Spyder, so I can read the values in LAB
> format (saved CGATS files).
>
>
>
> 3. White Point LAB L* is usually around 94, black point LAB L* is usually
> from 3 - 15, depends the type of media, so i need to slice this range to
> equal steps to find the linear values (if i want something else, not like a
> linear line, but a special curve, i need to slice this range according to
> that curves).
>
>
>
> 4. Normalize both LAB L* data arrays to a 0.0 - 100.0 range.
>
>
>
> 5. Find the difference between the measured ones and the linear ones.
>
>
>
> 6. Used those differences to calculate LAB L* -> Grayscale values. (This
> is the point where I think I misunderstand something). Normally I added
> these difference values to the original source Gray -> LAB L* values.
>
>
>
> I can use the corrected LAB L* values to create normal output profiles, or
> can use the Grayscale values to create a simple DeviceLink profile. The
> easiest method is to use the cmsBuildTabulatedToneCurve16 with the
> resulting values converted to the 0 - 65535 range. (But also got advice to
> use an 1D LUT table, but the ToneCurves are needs to be enough)
>
>
>
> Everything seems right and works flawlessly, except the results are not
> (very) linear and the  classic 'bullseye' images are wrong....
>
>
>
>
>
> From anyone who tried to do similar linearizations in the past or any hint
> or help would be very helpful for me.
>
>
>
> Many thanks,
>
> Tamas
>
>
>
>
>