[Bf-committers] Super-sampling and anti-aliasing

[tevhadij shwejpoi]

Dear all,

I posted the message below to the  Academic & Research list, where I
was advised to post here.

I am a computer graphics professor and one of my areas of expertise is
sampling. I have recently designed a family of new supersampling
approaches (better kernels) and found them to work very well in my own
prototype. The resulting images are sharper and less aliased than
those obtained with standard kernels such as box, tent,
mitchell-netravali, catmull-rom etc. The downside is that noise can
become more apparent if present.

The implementation is not particularly complicated, and it was a
simple matter to modify pbrt-v2 to support these new kernels. As the
next step, I decided to implement them in Blender, figuring that this
would benefit a larger audience. In the process, I have been somewhat
confused with what I can infer from Blender's source-code.

Is there a kind and soul reading this list who is interested in this
topic, knows a bit about the workings of Blender, and available to
exchange a few words with me? It seems as though it may be possible to
improve Blender's sampling even disregarding my own research efforts.

Kind regards,
Tev