W3C home > Mailing lists > Public > www-svg@w3.org > August 2012

Re: SVG Compositing

From: Rik Cabanier <cabanier@gmail.com>
Date: Sun, 5 Aug 2012 16:44:09 -0700
Message-ID: <CAGN7qDAsEA=h=ec-WzHKwT4eYjG6HrZG-_uJwP5GbG+me2FA4Q@mail.gmail.com>
To: Bob Holmes <rangsynth@gmail.com>
Cc: "Tab Atkins Jr." <jackalmage@gmail.com>, www-svg@w3.org
On Sun, Aug 5, 2012 at 4:25 PM, Bob Holmes <rangsynth@gmail.com> wrote:

> The only confusion remaining is just the difference between formulas.
>  The formulas in the new spec are all there for the basic porter duff
>  operators but are partial for the other pdf style blend modes like
>  colorburn.
>  But in the SVG document http://www.w3.org/TR/SVGCompositing/ the
>  formulas are all complete.

Correct. This was done intentionally.
The document shows how you can derive the full formula at the end of

Not all implementations may want to combine the formula either (ie
gpu acceleration will most likely not do so)

>  Also it is nice that the formulas in
> http://www.w3.org/TR/SVGCompositing/ are guaranteeing a premultipled
>  output, but there are cases for use in a canvas where it is beneficial
>  to have either premultipled or non-premultipled input and output.
>  I assume that for most imps it is enough to just describe the
>  premultiplied cases, like for the old
> http://www.w3.org/TR/SVGCompositing/ document.

>  To try and straighten out the confusion I experienced, it is between
>  how you draw once onto some pixels and then draw again onto the same
>  pixels.

Since a user never has access to the pixels, I don't think this is
How this is implemented is up the browser or SVG viewer.

>  The formulas in the new document describe using unpremultipled stuff
>  but they return a premultiplied value. If you plonk the result into
>  the pixel and then draw into the same pixel again you would need to
>  unpremultiply to run the formula again if reading it naively.

Where is that?
The document doesn't say how to implement this so it's again up to browser.

>  So the formulas in http://www.w3.org/TR/SVGCompositing/ are easier to
>  read as far as understanding the process for drawing like multiple
>  times onto the same place. It is just confusing to kind of loop the
>  result pixel back into the next formula as described in the new
>  document, and again the new document does not describe full formulas
>  like the other older document.
>  In addition, there is a place in the
> http://www.w3.org/TR/SVGCompositing/ document that describes the
>  difference between "Dca" and "Dc" but that is a long way from where
>  the formulas are stated nicely using the Dca approach.
>  So to try and communicate what my confusion was I guess it was the
>  notation and also the lack of full formulas for both premultipled and
>  unpremultipled cases.

premultiplied vs unpremultiplied is just:

cs = ás x Cs

>  And lastly how does a premutliplied result pixel get written into a 24
>  bit bitmap or a 1 byte bitmap for example. I read you say to matte it
>  with white like before displaying on a screen.
>  So what I did for a premultiplied case is to just assume that when
>  writing premultiplied results to 24 bit bitmaps is...
>    24Bit.Red = 32PM.Red + (255 - 32PM.Alpha)
>    24Bit.Blue = 32PM.Blue + (255 - 32PM.Alpha)
>    24Bit.Green = 32PM.Green + (255 - 32PM.Alpha)
>  This is same effect as matte it with white? Thus the color can then be
>  sent to the output with Bitblt or whatever the display requires.


>  So again. The confusion in the documents for me is a kind of lack of
>  description about how the model of drawing and then drawing again
>  works, and how it relates to underlying pixel formats such as taking a
>  premultiplied result to screen or to 24 bit bitmap goes.
>  But I understand it all now. Just maybe this info will help with your
>  work. Thanks very much.

Thank you!
I think we need to work a bit on the language that combines the blending
and compositing to make it less confusing.


> On 8/6/12, Rik Cabanier <cabanier@gmail.com> wrote:
> > Yes, please let us know if you find something in the latest spec
> confusing
> > or if you find a bug.
> > Also, if you spot differences (apart from combining blending and
> > compositing) let us know as well.
> >
> > Rik
> >
> > On Sun, Aug 5, 2012 at 3:30 PM, Tab Atkins Jr. <jackalmage@gmail.com>
> > wrote:
> >
> >> On Sun, Aug 5, 2012 at 2:46 AM, Bob Holmes <rangsynth@gmail.com> wrote:
> >> > There is an SVG document called "Compositing and Blending 1.0" dated
> >> > "25 July 2012" somewhere around and it splits the compositing and the
> >> > blending into two steps. That made for sense to me but now in the
> >> > other SVG spec with the "comp op" properties it puts the blenders and
> >> > the compositors into the single property, so therefore the result of a
> >> > multplication or overlay is not then subject to any further
> >> > compositing operator.
> >>
> >> The "Compositing and Blending" spec you mention
> >> <http://dvcs.w3.org/hg/FXTF/raw-file/tip/compositing/index.html> is
> >> the one you should be paying attention to.  Anything else is an old
> >> draft that has been superceded by the new one I just linked to.
> >>
> >> If there's anything confusing in the C&B draft, let us know, but you
> >> can safely ignore any confusion stemming from the old drafts.  ^_^
> >>
> >> ~TJ
> >>
> >>
> >
Received on Sunday, 5 August 2012 23:44:38 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 8 March 2017 09:47:29 UTC