Topic: Image base Spam emails

[mercyh]

Hi,

I have been watching this over on the contribs forum. Do you feel like this is ready for production?

Over the weekend with Spamassin RBLS enabled and bayese trained for 6 months I received 8 image based spam in my inbox. Another 12 hit my junkmail folder with scores between 5 and 10. I have about 60 users that don't even know what spamassin is so would like to be sure it is working before I implement.


Thanks for all your work.

Royce

[gregswallow]

> Do you feel like this is ready for production?

If it was, it would be in the smeupdates or smeupdates-testing repository.

It will be ready sooner if people test it and report bugs.  Upstream has a bug tracker now as well at http://fuzzyocr.org.

[cpuffalt]

John...

Corey,

How has this been working?

John

Well it hasn't blown up on me.  I did a quick grep through my junkmail folder and it scored a hit on about 5% of the emails there.  Of course that doesn't mean spamassassin wouldn't have caught many of those without FuzzyOCR.  It seems to have helped the epidemic though it hasn't cured the problem as a number of image-based spams are still getting through.

Keep in mind though that I'm just using this on my own personal server so the email volumes aren't very high.  I imagine FuzzyOCR would add considerable load to a busy site.  

I also have some issues with the current design of FuzzyOCR.  It relies on a fixed list of keywords.  It's too bad it doesn't/can't(?) leverage the existing bayesian-based scoring built into spamassassin so it could be more adaptive.

Anyhow, it's not a silver bullet but if you're running a low volume site or have a server with lots of headroom it may be worth installing... At least until a better solution appears.

Corey