For visvis, you might be interested in the different weight functions (I translated those of the antigrain library that are used in matplotlib):

http://code.google.com/p/glumpy/source/browse/glumpy/image/filter.py
http://code.google.com/p/glumpy/source/browse/glumpy/image/filter.vert
http://code.google.com/p/glumpy/source/browse/glumpy/image/filter.frag

You're right, a Gaussian filter is probably not the best choice for aa. I took a look at your filters and implemented a Lanczos filter (with dynamic bandwith depending on image scale). Looks much better because it does not blur so much. Thanks!

Almar