>>>>> "Paul" == Paul Rubin <"http://phr.cx"@NOSPAM.invalid> writes:
Paul> Steven D'Aprano <[EMAIL PROTECTED]> writes: >> Instead of displaying an obfuscated image of a nonsense word, display >> six randomly chosen photos, where five are of the same thing but not >> the same image. E.g. you might show five different kittens and a >> horse. The user has to click on the image that is not the same as the >> others. State-of-the-art horse-recognition software is not yet in >> widespread use by spammers *wink* Paul> No need to recognize the horse. Just pick one of the pictures at Paul> random and you'll get the right one 1/6th of the time. Repeat ad Paul> infinitum--they're spammers and like to repeat stuff after all. Paul> That's why those conventional captcha images make you recognize a Paul> multi-character string: so the guessing chance is low. Actually, the ones I saw that used a set of "one of these things is not like the other" images gave you a pop-up menu of maybe 100-200 words. The user needed to choose the name of the different object from that list. That makes it a bit harded to guess. Of course, these sorts of tests suffer from the same shortcoming as the randomly generated string. Visually impaired people have trouble with it. I finally settled on just reusing the SpamBayes engine to detect/reject spam submissions. Skip -- http://mail.python.org/mailman/listinfo/python-list