fwiw,
Some of the issues with merging HDR pixels have been hashed out in Nuke for a
while, so a peek at its Merge node may be helpful:
For example 'screen' in Nuke is implemented to avoid the 1-A negative problem:
if (A >0 && A<1 && B>0 && B<1) C = A+B - A*B
else if (A > B) C = A
else C = B
It also has the operation named 'hypot'(hypotenuse) or 'diagonal' which works
pretty well for HDR pixels:
C = sqrt(A*A + B*B)
And lastly 'geometric':
if (A>0 && B>0) C = 2*A*B / (A+B)
else C = 0
-jonathan
On Jul 13, 2012, at 9:37 AM, Jeremy Selan wrote:
> Yup.
>
> screen = A+B-AB, where AB term represents the if intersection of the
> subpixel geometry.
>
> But in the case where you have two mattes where you know the
> correlation == -1, (which denotes no overlap in the subpixel
> geometry), adding mattes reduces to a simple A+B. An example of no
> overlap in the subpixel geometry would be the alpha rasterization of
> two adjacent triangles in a mesh.
>
> So to summarize, if the subpixel geometry has a correlation of 0, the
> math for combining alphas is A+B-AB. If the correlation is -1, the
> math is A+B. And for perfectly correlated subpixel geometry? Sounds
> like a good white-board project for this weekend... Surely there's
> some paper from 1985 that has worked this all out? :)
>
> -- Jeremy
>
> On Thu, Jul 12, 2012 at 9:05 PM, Troy Sobotka <[email protected]> wrote:
>> And the inevitable question: What is the canonical compositing formula
>> for correlated then? Simple add?
>>
> _______________________________________________
> Oiio-dev mailing list
> [email protected]
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
_______________________________________________
Oiio-dev mailing list
[email protected]
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
