Milter-greylist

According to manu@...:
> 
> > What you can do is to feed all emails matching this to 'gocr' then check
> > the result against a list of words..
> > That may not be enough as direct ocr is still too weak (animated gifs,
> > transcoding, and stuff..). Add that and you obtain FuzzyOcr.
> 
> Not quite. In order to get FuzzyOcr, you still need ot have it written
> in Perl and with a dependency on spamassassin :-)

no, I said you can do it yourself using something that can play with
MIME parts, plenty of tools to manipulate pictures, an OCR of some kind,
and a dict-based rule.
Then, you will have something similar to FuzzyOcr. I don't think
it's wise to put that in milter-greylist, it should be yet-another-milter,
but IMHO, it's redundant. The bonus of doing that in spamassassin is scoring.

With FuzzyOcr, I got no false positive so far and very few misses,
so that's enough for me at the moment. I'm just concerned by the spams
that still get through, despite all what I already have in place.

[ my current problem is with valid mailing lists that let spams pass through.
greylist is useless here, and spamassassin is of no use because of bayes
and autowhitelist who think the from are good guys.. indeed, it's the list,
not the initial from. If I disable bayes (only 0-50 as this is the range
decreasing the score), then all yahoo groups are completely seen as spam,
and plenty of other valid stuffs too. I'm currently trying to adjust the
weigths of bayes instead of disabling them but I would prefer a better
option that I have yet to find. ]

/Fabien