I spend a fair bit of time manually shrinking icons then trying to
touch them up to look decent.

I figure this process aught to be automatable.

1. when shrinking it does antialiasing.

2. when shrinking it does antialiasing with background transparency
(png).

3. maybe if it were really clever it could try to extract regions of
the icon logically, then blow the icon up and smooth the regions and
sharpen the edge using spline curves, then start from that point to
shrink down again, rather than trying to match the precise pixel grid
of the original.

the idea might be you create icons on a large scale, perhaps starting
with photos, and let the software deal with much of the work of
dealing with low res pixelation, much the way PostScript fonts are
rendered.

has anyone tackled a project like this?