[Openicc] seeking for advice on rendering intent and black point compensation

[Craig Ringer]

Kai-Uwe Behrmann wrote:
> Hello,
> 
> Am 03.10.07, 12:47 +0200 schrieb Sven Neumann:
> 
>> When opening an image file that has a color profile embedded, GIMP
>> offers to convert from the embedded profile to the configured RGB
>> workspace profile (which will be sRGB for most users). This conversion
>> currently happens with hardcoded parameters for rendering intent
>> (perceptual) and black point compensation (off). My questions are:
>>
>> (1) Are these good default values or would you advice to change them?
> 
> For an simplistic sRGB workflow editing all in preceptual sRGB might be 
> considerable. 

Yep. Using sRGB minimises conversions since most inputs, and most
outputs targeted by the majority of GIMP users, will be sRGB. While one
of the wider gamut spaces is in many ways better for a working space,
those are only _really_ attractive when you have high precision pixel
values too. Given that, I'd agree with sRGB as a default working space
personally.

I do wonder if defaulting to converting mismatched images to the working
space is the best idea, though. I'm increasingly inclined to leave
images in their native colour space and let the colour management system
take care of making sure I get a reasonable view of the image on screen.

>> (2) The rendering intent used for the display color correction can be
>> configured in the Preferences dialog. Would it make sense to reuse this
>> setting for the conversion when loading an image? Doing so would make
>> sure that, with display color correction enabled, the image would look
>> the same whether it is converted on load or not. I am currently
>> inclinded to follow this argumentation and to change the code
>> accordingly.
> 
> Coupling the display and conversion rendering intent seems logical, and is 
> in fact deployed in other applications, namely CinePaint, Krita(?) ...

That would certainly match my expectations as a user. Getting a
different result on screen when converting an image to the working space
(vs using its native spaces as the working space for the image) would be
quite surprising.

> BPC is used in conjunction with relative colorimetric only. For printer 
> simulation, allowing different settings seems appropriate.
> Be warned about using relative colorimetric without BPC. Your users will 
> seldom expect such results.

I increasingly think that user interfaces need to offer _five_ rendering
intents, with a strong emphasis placed on the first two shown, eg:

Perceptual
Relative Colorimetric with BPC
-- Rarely Needed --
Relative Colorimetric without BPC
Absolute Colorimetric
Saturation

The "Use black point compensation" option would then be removed
entirely. (Or are there circumstances when it does make sense for other
intents?) Combined with a bit of documentation, focusing on the choice
between the first two, that might make it a lot easier for users to get
started.

--
Craig Ringer