In article <39E60BF7.A7CE2A8D@schunter.etc.tu-bs.de>, 
chr### [at] gmxde wrote:

> That looks promising, but if i understand things right both blurred 
> reflection and transparency are only fakes for rough surfaces (of 
> course that doesn't mean they are obsolete)

True, they are intended to simulate evenly rough surfaces. You can get 
more precise and controllable results using a normal, but you need 
antialiasing to get the effects to be visible, and that affects the 
whole scene and is slow.
If I can figure out a way to supersample specific textures or objects, I 
will try to implement it...it could use a method similar to the blur 
algorithms I am working on now. If I can work out how to get an estimate 
of the ray footprint, a sort of imitation differential ray-tracing might 
be possible.


> The structures in the second version seem to have some similarities 
> with shadows from area lights. 

Yes, because the rays "fan out" and are evenly spaced, the spacing gets 
larger with distance, similar to the way area shadows spread out.
These artifacts should be less visible with more random backgrounds(the 
checker pigment used here really shows the problem because of it's 
repetitive nature).
I suppose if the original algorithm is like media method 1, the current 
one is like media method 2. I have a couple ideas for anti-aliasing the 
blur:
1) Sample along 2 perpendicular directions, like the existing method 2, 
but super-sample between two samples when their difference in color 
exceeds a threshold. This would be sort of like media method 3...
2) Send rays out in a triangular pattern, dividing into sub-triangles 
when necessary. This would have the advantage of covering an area of 
space instead of sampling along two directions...the recursive triangle 
pattern might make the aliasing less noticeable, too.


> I estimate both algorithms take quite long to render compared to normal
> transparency depending much on what's behind the transparent object.

Compared to ordinary transparency: yes, it is much slower. However, the 
second version can sometimes produce smoother results with the same 
number of samples than the first version, so the slowdown isn't too bad, 
especially if it only covers a small area of the image.
These images both had 12 samples, and rendered in about the same 
time(around 2 minutes, though I haven't performed a real test of 
rendering speed yet).


> BTW, I remember you working on some pattern blurring function, did 
> you have any results in that direction ?

I am planning to make another try, the first one wasn't very successful 
or easy to use(it used a 3D convolution matrix, which allowed effects 
other than blur, but was a pain to use and too slow with matrices large 
enough to get decently smooth blur. I plan to allow this and another, 
easier to use and faster method in my next try.). I am debating whether 
to blur patterns or pigments...probably both.

-- 
Christopher James Huff
Personal: chr### [at] maccom, http://homepage.mac.com/chrishuff/
TAG: chr### [at] tagpovrayorg, http://tag.povray.org/

<><

Post a reply to this message