This explains why it's a bitmap:
http://www.25hoursaday.com/weblog/2007/08/21/FacebookTheSocialGraphRoachMotel.aspx

Appears to be Tahoma at 11px with font-color set to rgb(50,50,50). Tried rendering your email address using PIL but the kerning and anti-aliasing wasn't identical to your supplied image, although the shapes of the @ and other letters were pretty spot on.

I had the idea to pre-generate the various glyps (including two-character glyphs, if any), make it so the image is 16 pixels high (there's enough whitespace) and use two columns of pixels as a Uint32 (white=0, the rest=1) and then build a "scan trie" of sorts in Python and dump it using JSON. Would allow for a compact representation usable from JS, even if compactness isn't necessary.

Facebook runs on PHP which uses GD for image generation IIRC, but I haven't got it installed to see if it renders things differently. One could, of course, collect a bunch of email images and run the test off that, but it's not as easy to automate things then.

Maybe I'm just over-thinking it, haven't eaten dinner... Does sound like a fun little project, though.

I tried mucking around with it for a bit, but I was getting a null return for getImageData every time.

I believe that was added after some were scraping email addresses. Since most people befriend anyone who asks, it's easy to gain a nice amount of data. Email address being the most valuable. Not such a bad move considering their business model is largely to not fall into the privacy trap that MySpace had. So far they have done pretty well.

Oooh and if someone could write a captcha solving algorithm too :-) it's got to the point where I can't read them anymore.

monk.e.boy

I have a truly marvelous solution to this problem which this comment box is too narrow to contain. I'll put it up on my blog shortly.

Here's the writeup of my solution: http://www.chrisfinke.com/2007/09/11/convert-facebook-e-mail-images-to-actual-e-mail-links/

Definitely a fun problem to solve.

Jolly clever. Nicely done.

You could register a click handler for the images and do the conversion on-demand, to avoid slowing down any page loads. And possibly add a pref for choosing between on-demand and on-load operation. For version 1.1 or whatever ;).

Chris: Most impressive! Thank you very much indeed. I wonder if they'll try and break the technique. I hope not.

Fredrik's idea is possibly a good one; I guess it depends on what the performance impact is. Can you use document.images to speed things up, if you aren't already?

If Chris' solution can do general image-to-text conversions, it would be extremely useful to add to the spam detection in Thunderbird. You wouldn't even need to get the conversion exactly right; e.g. just finding "Vioxx" or other keywords in the image would be enough to flag the email as likely being SPAM.

@monk.e.boy

"Oooh and if someone could write a captcha solving algorithm too :-) it's got to the point where I can't read them anymore."

It's been done. One of the members of theReformed (http://www.thereformed.org) wrote one and published it about a year ago.