Re: [Fwd: Pre-multiplication of samples in Filter Effects]

Raph Levien wrote:

> I would disagree that, in general, premultiplied alpha is a
> significant loss of precision. Yes, when converting back to separate
> alpha in the small-alpha case, roundoff errors get magnified. But when
> you take that pixel and composite it over another pixel, you only get
> an LSB or so of roundoff errors.

Right. So, it depends on where you measure your precision - on one
layer, r on the total composited result.

> For Web work, that should be completely satisfactory.
> 
> Now, if we're talking about high-quality print applications, then 8
> bits is simply not adequate. 

For decent quality, even online, 8 bits is not suficient (rtemember, its
not just premultiplying - its also going from a partially -perceptual
space (because of 2.2 gamma correction) into a linear light space.

> Going to 16 bits, premultiplied or no,
> is the only way to go. 

I agree, both for screen and print, for good quality. Then again, lots
of folk do good work in tools like Photoshop, which only really supports
8 bits per component (you can move to 16, but then you don't get to do
anything much except play with curves).

> My take on the draft is that it's not being
> targeted for high-quality printing. 

Huh? Where does it say "and please keep all intermediate results at 8
bits per component so it looks crappy" ? ;-)


> P.S. By way of introduction, I'm one of the Gimp developers and author
> of the antialiased renderer for the Gnome canvas. I'm planning on
> doing a full-fledged SVG implementation in the Gnome framework.

Hi, and welcome!

SVG, for its target market, really requires antialiased rendering. 

Please let this list know as soon as you have any more information (eg a
web page) about your SVG renderer, and I will link it in to the public
SVG overview page on the W3C site.

--
Chris

Received on Friday, 4 June 1999 11:25:58 UTC