i like this idea, possibley even design a DCC / pyzor / razor type system which would implement a distributed checksum / md5 etc on all image ?
----- Original Message ----- From: "Manuel Schmitt" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Sent: Thursday, January 15, 2004 4:08 PM Subject: [SAtalk] Improvement: Image Recognition as spam criteria > Dear readers, > > while using Spamassassin for about one month and having a very good > recognition rate I am discovering that spam that has almost or no text > within does not get detected by SpamAssassin, neither by normal criteria nor > by the Bayes filter. I think because there is not enough information for the > Bayes filter to be able on a reliable decision. > > Now I had the following thought. What about a special image database which > is maintained in a Bayes-like-style. I could imagine the following. While > learning spam with sa-learn, the learning process could also filter out each > image in an email. Then we have to build a more unique representation of > this image, comparable with a hash value of a string. Then we store this > "hash value" as entry a Bayes-like image database. When new email comes in > we do a comparison of all images in this email with them in our database. > > I don't have any idea of what resouces this addition check(s) would consume, > but perhaps it would be a nice addition feature. I was concerned about some > ways how to get a hash value out of an image, if someone is interested, feel > free to contact me. > > Best regards, > Manuel Schmitt ------------------------------------------------------- This SF.net email is sponsored by: Perforce Software. Perforce is the Fast Software Configuration Management System offering advanced branching capabilities and atomic changes on 50+ platforms. Free Eval! http://www.perforce.com/perforce/loadprog.html _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
