Re: [Gimp-developer] The new "linear" and "gamma" precision choices

[Michael Henning <drawoc darkrefraction com>]

Elle: Things are confusing now because the code isn't finished. A lot
of things will change before 2.10 is released.

Rest assured, we won't keep everything this way.

On Sun, Jul 7, 2013 at 5:23 PM, Elle Stone <l elle stone gmail com> wrote:
> Gimp from git allows the user to choose which image precision to use
> while editing the image. The choices used to be 8-bit, 16-bit, and
> 32-bit integer, plus 16-bit and 32-bit floating point. The choices
> have recently doubled: for each image precision choice, the user must
> choose whether to work in a "linear" or a "gamma" color space. The
> default seems to be "gamma" rather than "linear".
>
> I did some experimenting to see what these new "precision" choices might mean:
>
> The practical result of choosing "linear" vs "gamma" as your image
> "precision" depends on the image's actual RGB color space TRC (whether
> linear or true sRGB TRC or some other TRC) and it also depends on
> which operations you want to perform. Limited testing suggests that at
> present it affects blend modes but not blurring or gradients.
>
> CASE A: the image is in the regular sRGB color space
>
> If the image is already in the regular sRGB color space, then choosing
> "gamma" means for example that a soft-light blend of an image over
> itself will look like it was done in the regular sRGB color space. But
> choosing "linear" means it will look like the blend was done in a
> linear gamma version of the sRGB color space and the result will be
> darker than it would be if done in the regular sRGB color space. In
> both cases gradients look like regular sRGB gradients. And in both
> cases, blurring looks like it was done in a linear gamma version of
> the sRGB color space.
>
> CASE B: the image in a linear gamma version of the sRGB color space
>
> If the image is already in a linear version of the sRGB color space,
> then choosing "gamma" means that a soft-light blend of an image over
> itself will look like it was done in a *linear* gamma version of the
> sRGB color space. And choosing "linear" means it will look like the
> blend was done in a gamma=0.45 (approx) version of the sRGB color
> space. In both cases gradients look like linear gamma sRGB gradients.
> And in both cases blurring looks like it was done in a gamma=0.45
> version of the sRGB color space.
>
> CASE C: the image is in some color space other than sRGB
>
> Some color management background:
>
> The "gamma" precision refers to the regular sRGB matrix color space
> tone response curve. However, the sRGB matrix color space is not a
> true gamma color space. Its tone response curve (TRC) is composed of a
> linear section at the toe and a true gamma=2.4 section everywhere
> else.
>
> A linear gamma version of the sRGB color space is the regular sRGB
> color space with the regular sRGB TRC replaced with a  gamma=1.0 TRC.
>
> RGB matrix working spaces like sRGB or CIE-RGB or a ProPhoto-like
> color space, etc, are defined primarily by three things: their RGB
> Primaries, their Tone Response Curve, and their white point. Putting
> white point aside (lcms takes care of that behind the scenes), a color
> space's Primaries determine the reddest red, bluest blue, and greenest
> green that color space can contain. A color space's TRC determines how
> fast a color goes from dark to light as the RGB values go from 0 to
> max white (255 for 8-bit integer, 65535 for 16-bit integer, 1.0 for
> floating point).
>
> So linear sRGB is not the only linear gamma color space. You can have
> linear gamma versions of any working space. And you can also have, for
> example, a color space with the CIE-RGB primaries and the regular sRGB
> tone response curve. ProPhoto-like color spaces usually have a
> gamma=1.8 TRC but you can create a linear gamma version of a
> ProPhoto-like color space, and a version with the regular sRGB TRC.
> And so on.
>
> What happens if your image happens to be in a color space other than
> the regular sRGB color space?
>
> The Gimp "gamma" and "linear" precision code assumes the image has the
> regular sRGB TRC. The closer a color space's TRC is to the regular
> sRGB TRC, the more the resulting tonalities will be like case A above.
> The closer the color space's TRC is to being linear, the more the
> result's will be like CASE B above. And results will be more or less
> wrong to the extent that the color space's TRC deviates from the sRGB
> TRC.
>
> May I ask what is the point or reason for this new and confusing
> option of "linear" and "gamma" precision? I say "confusing" because it
> changes the mathematically expected behavior given the TRC of the
> image color space.
>
> I should probably confess that for my own image editing I use a
> version of babl/gegl/Gimp where the babl conversions between linear
> and regular sRGB are eliminated. That way I can choose which color
> space I want to work in and be confident of the results: linear gamma
> working space, linear gamma results; higher gamma working space,
> higher gamma results. I fear this "out" will not be viable for long.
>
> Elle, as usual, puzzled
> _______________________________________________
> gimp-developer-list mailing list
> List address:    gimp-developer-list gnome org
> List membership: https://mail.gnome.org/mailman/listinfo/gimp-developer-list