On occasion, I've done that as well-- downsizing large renders in Photoshop,
using bicubic interpolation there--with pretty good results (compared to using
POV-Ray's 'default' AA with its settings of type 1, threshold of 0.3, and
jittering on.) But I don't use AA's default settings, as they are somewhat
crude, visually speaking.

Your question prompted me to do some more research, mainly about Photoshop's
various methods of downsampling-- bicubic interpolation vs. bilinear vs.
"nearest neighbor."

I looked at 'bicubic interpolation' and 'image scaling' at Wikipedia; the latter
leads me to believe that bicubic and bilinear methods may apply a slight bit of
useful blurring before the downsizing-- or rather, some low-pass filtering, to
help minimize the 'high-frequency' sharp color/brightness transitions between
pixels. That sounds a bit like AA, although not quite the same thing.

I just did a test with POV-Ray's included 'biscuit' scene file-- first rendering
it at 3200 X 2400 pixels with no AA, then again at 800 X 600 *with* AA (using
the defaults.) I then downsized the larger image, first using bicubic
interpolation, then bilinear. Comparing the resulting images close-up in
Photoshop, I actually prefer the bilinear method of interpolation-- it seems to
'match' the look of the smaller AA image better, regarding fine details, which
are retained (more or less.) Bicubic looks a bit *too* smooth--an interesting
discovery.

BTW, something I read long ago (and my own intuition) tells me that, when
downsizing an image, it's best not to use a straight 4-to-1 reduction (as in my
example), or a likewise purely linear reduction. In other words, to choose
something different (odd?), like 7-to-1 or 3-to-1, etc. If I recall correctly,
the interpolation works better this way(?), visually.

Post a reply to this message