2006.09.16 22:00 "Re: [Tiff] is there alpha component present in GrayscaleorPalettecolorimage", by Toby Thain
On 16-Sep-06, at 5:50 PM, Joris wrote:
PNG does not support associated (pre-multiplied) alpha. It only supports unassociated (simple) alpha. In this case, supporting alpha from within the colormap makes sense. For example, you may assign one color to represent all completely transparent areas (like GIF). TIFF does support unassociated (simple) alpha but the TIFF specification does not define how it should be applied to the image to obtain a viewable result.
I have another take on this. The difference between pre-multiplied (associated) and ordinarry non-multiplied alpha, is merely a multiplication (except when alpha indicates complete transparency). That's just a maths operation. So in my mind, the fundamental quirck about having alpha inside palette entries is independent of whether you code it as pre-multiplied or non-multiplied. The simple fact remains you're making fixed associations between colors and alpha, as opposed to coding alpha as per actual pixel position. It's this simple fact that makes a huge impact on the code required. For example, dithering from full-color-with-alpha to palette-including-alpha is the challenge,
Why is that challenging? You just omit that index from colour matching consideration. In most cases, just thresholding alpha to that index would be sufficient for opacity handling.
> whether those palette entries are coded pre-multiplied or non-
doesn't have an impact in this regard.