I was kind of interested in doing this for Unicode a while back, which has the potential to provide for a lot more possible characters and thus a lot more-accurate renditions. If you don't care about real-time operation, I suspect that you can do this with off-the-shelf software by just writing a small amount of code to generate an image for each character
a shell script driving ImageMagick could do it
and then feeding it into photomosaic software, like metapixel.
The major limitation is that unless you're just interested in doing this for the text-based aesthetic and are actually rendering and presenting an image to the end user
think something like Effulgence RPG, Warsim, Armoured Commander II, Armoured Commander, Cogmind, SanctuaryRPG, Cataclysm: Dark Days Ahead, Stone Story RPG, Roots of Harmony, and so forth
you can't control the font that the thing is rendered in on the end user's computer. And the accuracy of the rendering degrades the more the typeface used on an end user's computer differs from your own.
It'd probably be possible to build some kind of system that does take into account the differences for different typefaces, scores characters higher based on checking for character similarity across different typefaces.
Note that there are also at least two existing libraries out there
what I can think of off the top of my head
that will do image-to-ASCII conversion
aalib and libcaca, the latter of which has color support. I also posted a tiny program some time back to generate images using the colored Unicode tiles, and I imagine that someone out there probably has a website that does the same thing.