Gerhard Fuernkranz wrote:
Another simple method for improving the accuracy for in-gamut colors in conjunction with a CIELAB-based B2A table might be the use of sigmoid-shaped prelinearization tables for the a* and b* channels, such that the CLUT operates on a distortet Lab space, while it still maps the complete Lab space to device space, giving a higher accuracy for near neutral colors (in-gamut) at the cost of a lower accuracy for highly cromatic
Something less subtle, like having two slopes with a "knee" might work just as well, as long as the "knee" was outside the device gamut.
And of course, since the B2A table is only 3D, while a CMYK A2B table is 4D, there is also more room for using an even finer grid resolution for the B2A tables (than for A2B), which also may help a bit to compensate the deficiencies of the B2A tables (i.e. the "wasted" space).
This is already the case to some degree. -qh 4D->3D is 17^4, 3D->4D is 33^4.
I'm still trying to find ways to see what's going on inside the B2A table. Is the current result as good as can be achieved (without using tricks involving the matrix or curves), or is there something peculiar going on ?