[Lcms-user] Re: LCMS 1.07b slow?

[<ma...@li...>]

Hi,

> Marti,
>  One more question.  I'm trying to figure out the proper coding for
> TYPE_XYZ_16.  I'm taking Kodak Q60 data and trying to feed the XYZ =
values to
> LCMS, but I don't seem to have the magnitude quite right.
>=20
> For example, on my Q60 color chart, white (Dmin) is XYZ: 77.49, 80.31, =
69.06
>=20
> What is LCMS expecting to see for these XYZ values?  For example, the =
Y value
> is 80.31.  I know that LCMS expects to see a word (16 bit number), but =
what is
> the value for 80.31?  I've tried these equations and neither seem to =
give the
> right result:

lcms does expect values between 0 and 1.99996. The function for convert =
from
floating point to fixed point is:

WORD XYZ2Fix(double d)
{
       return (WORD) ((Fixed32) (DOUBLE_TO_FIXED(d) >> 1));
}

In Delphi this is something like

    (round(x * 65536.0)) shr 1

XYZ does operate on 1.15 fixed point, so, doing * 65536 does convert to=20
15.16 and the shift right does convert to 1.15.

About the 80.31, well, what happens is that IT8 does use "absolute" =
values.=20
Just divide by 100 before passing to this function.

Taking your data,

80.31 -> Divide by 10 ->  0.8031 -> Apply XYZ2Fix -> 26316

Besides, If you  plan to use XYZ, be shure to include =
cmsFLAGS_NOTPRECALC
in the "flags" parameter of cmsCreateTransform(). This is needed since =
trilinear
interpolation could give slightly wrong numbers... XYZ does operate on =
gamma 1.0,
and the gamut is so big that interpolation would need twice as points as =
it is currently=20
using.

Regards,
Mart=ED Maria
The little cms project
http://www.littlecms.com
ma...@li...