New Ways to Verify Human Input

Experimental techniques are hard to break

[rogerd]

Sites like Overture and AltaVista, among many others, have added verification steps to try to prevent access by automated programs. Mostly, these take the form of obfuscated characters that are supposed to be difficult to OCR.

A visit to [captcha.net...] (a project at Carnegie-Mellon U) shows some of the new stuff being tested. They came up with a really tough OCR challenge, with words almost completely overprinting other words. As challenging as this is, it seems that other labs have already broken this and are achieving a fairly high success rate with automated reads.

Some of the other techniques involve requiring the user to look at photos of animals and then identify the types, or to identify a common item in several different photos.

One, dubbed Bongo, requires the user to match graphic elements with one of two groups of other elements; the first one I viewed was not trivial. I don't know how effective this is against rogue programs, but I think that it might be right for your site if you share a lot of traffic with, say, mensa.org. ;)

[jeremy goodrich]

Very interesting research going on in that area - makes me wonder:

If the image parsers are so advanced, can they now tell the difference between 'auto generated' text & hand written?