Re: [Gimp-developer] (no subject)

[Øyvind Kolås <pippin gimp org>]

On Tue, Jun 19, 2012 at 3:24 PM, Elle Stone <l elle stone gmail com> wrote:
> (1) Color space conversions result in lost colors, whether done at 8-,
> 16-, or 32-bits, floating or integer. At first I thought it was a
> result of converting to and from a linear-gamma color space. But
> actually the color space gamma makes no difference.

> (2) All the color space conversions that I used involved matrix
> profiles. For a matrix profile, the only valid type of color space
> conversion is "colorimetric". I used "relative colorimetric" color
> space conversion exclusively. The color space conversion results
> varied depending on whether I used black point compensation or not.
> However, if the color spaces involved all have the same black point,
> using or not using black point compensation should not make any
> difference whatsoever.

None of the ICC profile handling in GIMP deals with GEGL or higher
bitdepths - at the start and end of all these code paths things go
through 8bpc; and probably behave similarly to how they do in 2.8 -
there shouldn't be much of a difference.

> (3) Eric Brasseur's image scaling test produced very odd image
> discolorations, regardless of what color space (linear or nonlinear
> gamma) or image depth (8-bit integer or 32-bit floating) was used.

I've seen some of these discolorations; but at least the results are a
lot more correct than earlier versions, for some of these tests the
type of resampler and "where in the pixel" the center being resampled
is will matter a lot, the 2x2 px in input -> 1 px in output hardcoded
tests used by eric brasseur are not generic like the code paths
employed by gimp.

> (4) Although Gimp 2.9 correctly opens 8-bit tiffs, when opening any
> 16-bit tiff, the image is subjected to a roughly gamma=2.2 curve. All
> RGB colors are altered (it's not just a display problem), so the
> resulting image is too light.

TIFFs seem to be a nightmare and only brief initial attempts have been
made at making GIMP support higher bitdepth diffs.

/Øyvind K.
-- 
«The future is already here. It's just not very evenly distributed»
                                                 -- William Gibson
http://pippin.gimp.org/ ;                           http://ffii.org/