Sunday, January 15, 2006

Do spammers fear OCR?

Nick FitzGerald recently sent me two sample spams that seem to indicate that some spammers fear that using images in place of words isn't enough. They've started to obscure their messages to prevent optical character recognition.

The first spam appears to be a scan of a document that's been skewed slightly. Now this could be a simple and bad scan.



But the second is even more interesting. It appears to be perfectly normal:



Until you look at the fact that this was actually constructed using <DIV> tags for layout and the breaks between the lines are in the middle of words. Here Nick has kindly inserted borders showing that the words are broken horizontally and then put back in the right position:



But is anyone doing OCR, or are spam filters getting good enough that the spammers are being really paranoid about what they are sending?

The funny think about the second example is that the URL they include is not obfuscated, is clickable and appears in the SURBL :-) So despite the effort to obscure the content a simple check of the spamminess of the URL gets this email canned.

6 comments:

Justin Mason said...

That *is* interesting!

Typically, when I've seen something like this before, it indicates that one of the proprietary services (AOL? GMail?) is indeed using OCR, and I just didn't know it yet -- but the spammers do.

John Graham-Cumming said...

Nick also mentioned another, non-OCR, possibility for doing this... if the images are skewed or split in different ways over time then the spammer prevents a spam filter from being able to track a common hash for each image.

mr g said...

On the other hand, most e-mail clients (even webmail services) do not show images by default, so what's the point of using images?

If I get such an e-mail, it'll probably land in my Junk E-mail. If it does not, it will be a blank e-mail with image placeholders.

Or am I missing something here?

Nick FitzGerald said...

"... most e-mail clients (even webmail services) do not show images by default, so what's the point of using images? ... Or am I missing something here?" -- mr g

I think what you missed is that, by default, the _most used_ Email clients do show inline HTML images from MHTML format messages that use the cid: protocol to reference those images. Perhaps only 80% of all potential recipients and only 7% of potential recipients who are *nix users will see the spam's graphical ad, but that really does not matter to the spammer, and matters even less if this new approach means that each week s/he gets a few hundred million more ads past the filters until they are further tweaked.

Of course we don't know if this is more successful or not, BUT the spam-trap address that snagged both the samples John referred to is still getting a goodly amount of the "skewed scan" style of "junk shares" spam so maybe that tells us something?

Nick FitzGerald said...

"But is anyone doing OCR, or are spam filters getting good enough that the spammers are being really paranoid about what they are sending?" -- jgc

Well, back in January, I think the answer was no-one was doing OCR, although there had been some talk about it. However, on 14 April 2006, Martin Blapp posted a link to his "Image validator/OCR SA plugin" for Spamassassin, to the MIMEDefang (and other??) lists. An archived copy of that mesage is at:

http://lists.roaringpenguin.com/pipermail/mimedefang/2006-April/029999.html

Sorin said...

They went even deeper in this anti-ocr issue. Now they send emails with an image inside which is created from multiple parts. The parts seem to be created with a generator.

See an email here:
http://www.mustaca.de/blog