[Openicc] printer, driver, CUPS, PPD, printing GUI, ICC-profiles, colord, Oyranos, taxi....

[Michael Sweet]

On Jan 30, 2012, at 12:09 AM, Graeme Gill wrote:
> Michael Sweet wrote:
> 
>> I know I have
>> been discouraged that ICC tables are based on a linear grid which requires a lot of grid points
>> instead of having most of them "near the edges" where the higher sampling is needed.
> 
> I'm not sure what you mean. ICC profiles generally use the per-channel input
> tables to distribute the grid in some "optimal" fashion.

Even with the input and output tables (which do help), I have always found issues when you start mixing a lot of ink - it is too easy to get "mud" in dark areas particularly, and then are the transition points between similar colors (e.g. from light cyan to cyan) that can be problematic as well - the per-channel input and output tables don't help much here, and the traditional trick of doing the mapping of a particular color (e.g. cyan) to the different device colors has its own issues since maintaining that transition and still providing density controls for media calibration can be tricky.

> [Your assumption about the edges is not necessarily true either
> - see "Improved Perceptual Uniformity of Sampling in Color Look-up
> Tables" by Rohit A. Patil and Maxim W. Derhak, pp 238, 18th Color Imaging
> Conference, 2010]


I can't find a free copy of that, but based on the abstract it sounds like they are doing an input mapping that effectively just makes the 3D LUT grid non-linear - a standard trick that still has limitations.

What I really want is a sparse grid representation - areas of the grid that have a relatively linear transition between adjacent points could be omitted and calculated (as needed) at profile load/use time to reduce the size of the profile itself.

________________________________________________________________________
Michael Sweet, Senior Printing System Engineer, PWG Chair