Re: [Openexr-devel] Color Management Proposal / Siggraph meeting

[Florian Kainz]

I guess I should have been more careful about the wording of the
section on painting:

In the proposal, we suggest that all OpenEXR images live in the color
space that we call scene-referred.  Several people had asked how one
would paint a scene-referred image.  One possibility is to make the
paint software aware of the proposed color management scheme.  In this
case all image manipulations happen in scene-referred space; in order
to display the image, the paint program continuously applies the SROM,
OMRD and RDPD.

The example where a primitive paint program saves images in a file
format where pixel values correspond more or less directly to screen
frame buffer values was meant to illustrate that this case can handled
in our scheme:  Even if all you have is an image that lives in the
display's color space, you can derive a scene-referred image from it.
The example is not meant as an attack on Photoshop or on any other
software with built-in ICC profile support.

Our color management proposal is meant to work in a film and video
production environment, where ICC profiles have, for whatever reason,
not been successful.  I don't really care whether the concepts described
in the proposal describes are a subset or a superset of ICC profiles
(I think they are neither).  The proposal attempts to address problems
specific to film production.

Florian


Chris Cox wrote:
> A few notes:
>
> A paint program does not necessarily paint directly into the display 
> framebuffer - it paints into an offscreen buffer and transforms the 
> color data to the display framebuffer.  Thus the painting may exist in a 
> colorspace totally unrelated to the display device(s) used.
>
> So paintings should have an image colorspace, with no reference to the 
> display or framebuffer.
> The inverse display transform should not be involved unless you are 
> dealing with a primitive paint system that does draw directly to the 
> framebuffer (at which point you simply use the display profile for the 
> image colorspace).
>
>
>
> There is already a well known and well used framework for describing 
> color transformations: ICC profiles. They do have limitations with 
> regard to HDR data - but that could be fixed.  Defining yet another 
> color transformation syntax seems a bit silly.
>
> Yes, the concept you describe is sound:  because it is a subset of the 
> well proven concepts already in use within the ICC color management 
> framework available in many professional applications and implemented in 
> common desktop OSes.
>
> Chris