Hi, > So it's not a platform difference, but a rounding error that affects > all platforms? For 8-bit editing, the error makes no difference. But > for 16-bit editing, 0.0003162277 times 65535 is 20.7240. So it looks > like at 16-bit integer, errors up to nearly 21 can be expected when > converting zero and near-zero channel values between different ICC > profiles, yes?
No, 16 bit have 16 bit of precision, 32 bit floating point have 7 digit precision and a wide domain, that is, you have +/-0.000001 but you also have the exponent. The range is huge and this is useful for things like raw photo where intensity varies broadly, and also have the negative numbers. But still, 16 bits gives you 1/65535 which is 0.00001 and just one order of magnitude below. If you want to go always from 0 to 1.0 with no negatives and no highlights, maybe 16 bits is more adequate. On float you can have an epsilon of +-0.000001 and this means zero can be slightly negative. I did some investigation on how to better the float precision. I think it can be done in some cases, for example in transforms holding only matrix to matrix, both matrices can be multiplied and this would avoid the quantization in the middle. This would greatly enhance some kind of transforms, but still the main problem is in CLUT based transforms which involves profile elements using floating point extension. This is quite uncommon as per today. In November, at DevCon we at the ICC are going to announce iccMAX, what would correspond to ICC V5. This new spec will handle floating point area more in depth. Right now in V4 is a sort of experimental thing that IMHO is not very well solved. The degree lcms will support iccMAX is still an unknown. http://www.color.org/DevCon/ But of course those are long term plans. Best regards Marti ------------------------------------------------------------------------------ Want excitement? Manually upgrade your production database. When you want reliability, choose Perforce Perforce version control. Predictably reliable. http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user
