[Bf-committers] Color space and alpha issues

Troy Sobotka troy.sobotka at gmail.com
Fri Dec 16 06:15:52 CET 2011

On Dec 15, 2011 1:24 PM, "Jason van Gumster" <jason at handturkeystudios.com>

> I understand the necessity/convenience of internally choosing one alpha
> over the other. However, I'm curious how this would impact saving to image
> formats that only support one alpha format (e.g. premultiplied alpha is
> explicitly not part of the PNG spec[1]). Granted, I could be missing
> and it's not a huge issue, but I figured it'd be worth asking.

A library like OIIO enforces an internal premultiplied format. All output
is then subject to a divide for formats that require unassociated alpha.
When doing an operation at 32 bit depth, the loss is considered negligible
with respect to the gains of an internally consistent format. As noted by
Brecht, the divide is mandatory for moving back and forth between
linearization and non linear values. It would also be required to
unassociate the alpha for any alpha channel manipulation such as math work.

Bear in mind that most industrial grade formats permit either form of
alpha, such as EXR.

With respect,

More information about the Bf-committers mailing list