2003.09.24 08:39 "[Tiff] [Lcms-user] Lab tiff with alpha", by Kai-Uwe Behrmann

2003.09.24 08:39 "[Tiff] [Lcms-user] Lab tiff with alpha", by Kai-Uwe Behrmann


I like to convert some tiff images from RGB to CIE Lab color space using lcms. Now they contain alpha and I run into trouble. As this is related to tiff and color management I like to crosspost.

How is alpha to been handled in CIE Lab tiffs?

The TIFF-6.0 spec presumes in section 18 to handle alpha allways as premultiplied (associated) alpha.

This makes sense for RGB values, since all channels include luminance. Lab has its luminance stored only in the first - the *L channel. If I premultiply channel *a and *b of the Lab data with alpha, I get strange color casts. Clear, these channels include no luminance and shall be per definitin luminance- and alpha independent.

The TIFF-6.0 is not clear to me how to handle alpha for Lab.

One thing I thought was to premultiply only *L with alpha (not TIFF-6.0 conform). Then an application needs to know how to handle an associated alpha channel depending on the color space - RGB other than Lab. This seems complicated to me.

Would it be better to handle associated alpha only with those color spaces which include luminance in all three color channels? How can alpha be handled in the case there symetric premutliplying is not possible like for CIE Lab? - as unassociated alpha?

I know this matter is far away from dayly used RGB, but Lab is requested more and more, and I want to produce exchangeable files.


Kai-Uwe Behrmann