Please pardon me if somebody has already beaten this to death, but
an idea for anti-aliasing occurs to me.

It occurs to me that when a ray-intersection test takes place, the
instances where the test nearly succeeds, or nearly fails, can be
determined from the numbers comapared during the test.  If I am
correct, for instance, a b^2-4*a*c test is made for quadratics,
the values of b^2 and 4*a*c will be very close during the cases I
am drawing attention to, meaning that if a cone of rays covering
the pixel had been shot, instead of the one ray, some would have
hit the object and some would have missed, and more inportantly,
the relative proportions of these two groups of rays can be
calculated from the exact results of the ray-intersection test;
for instance, if b^2 exactly equals 4*a*c, we can say that the
ray intersects the surface at one point, and is tangential there
as well, and therefore half of the rays bundled around the tested
ray would miss, and the others hit.  If b^2 is slightly greater
than 4*a*c, then a greater number hit than missed, and so on.

This in turn leads to the idea that if this proportion is calculable,
the ray could be treated as if it had intersected the surface in
question, and another ray shot, ignoring the object in question,
and its results averaged with the intersecting ray, to give an
anti-aliased result.

It already sounds a bit complicated, but I just wanted to throw
the idea out there.

Regards,
John
-- 
ICQ: 46085459

Post a reply to this message

As far as I understand it you are talking about an anti-aliasing
scheme which would save tracing unnecessary rays near object edges.
How about anti-aliasing textures, shadows (can shadow rays benefit
from similar optimisations?), etc.?


Peter Popov ICQ : 15002700
Personal e-mail : pet### [at] usanet
TAG      e-mail : pet### [at] tagpovrayorg

Post a reply to this message

On Sun, 09 Apr 2000 15:38:38 -0400, John VanSickle wrote:
>It occurs to me that when a ray-intersection test takes place, the
>instances where the test nearly succeeds, or nearly fails, can be
>determined from the numbers comapared during the test.  If I am
>correct, for instance, a b^2-4*a*c test is made for quadratics,
>the values of b^2 and 4*a*c will be very close during the cases I
>am drawing attention to, meaning that if a cone of rays covering
>the pixel had been shot, instead of the one ray, some would have
>hit the object and some would have missed, and more inportantly,
>the relative proportions of these two groups of rays can be
>calculated from the exact results of the ray-intersection test;
>for instance, if b^2 exactly equals 4*a*c, we can say that the
>ray intersects the surface at one point, and is tangential there
>as well, and therefore half of the rays bundled around the tested
>ray would miss, and the others hit. 

This is true only if the bundled rays cover an area much smaller than
the object. Assume that the object covers only part of a pixel, and
touches the center of the pixel, like this:

  +------+
  |    _ |
  |   / \|
  |   \_/|
  |      |
  +------+

A ray shot down the middle of the pixel touches the object
tangentially, and the test above yeilds zero, but the object covers
less than half of the pixel. 

I am not sure if this makes much of a difference in the final picture
(especially since the object would probably vanish completely if moved a
bit to the right), and it might help in determining whether antialiasing
should be used on this pixel.

	hp

-- 

|_|_) | Sysadmin WSR     \ Durchbruch in der Bionik, und Microsoft geht
| |   | hjp### [at] wsracat     \ Pleite und Gardena bringt organische PC's
__/   | http://www.hjp.at/ \ auf den Markt.           -- Stefan Schaefer

Post a reply to this message

You forget that the surface may have a texture which needs to be
antialiased as well. We have to shoot all those rays in order to get the
average color of the surface.
  We also have to shoot all the rays not intersecting the surface in order
to get the average color of what is behind the surface.

-- 
main(i,_){for(_?--i,main(i+2,"FhhQHFIJD|FQTITFN]zRFHhhTBFHhhTBFysdB"[i]
):5;i&&_>1;printf("%s",_-70?_&1?"[]":" ":(_=0,"\n")),_/=2);} /*- Warp -*/

Post a reply to this message

Peter Popov wrote:
> 
> As far as I understand it you are talking about an anti-aliasing
> scheme which would save tracing unnecessary rays near object edges.
> How about anti-aliasing textures, shadows (can shadow rays benefit
> from similar optimisations?), etc.?

Actually, using it for shadows might ben better than using it for
anti-aliasing.

-- 
ICQ: 46085459

Post a reply to this message