问题:

I do the web page for my local library, and I was thinking it might be kind of appealing to have a "custom" captcha based on book covers. So serve up one of several dozen book covers, and have the patron filling out the form type the book title to prove they're human. Assuming I stripped the title/author info from the image and filename, would that be enough? Would the fact that it was a unique system on a fairly small website be enough to make it effective? Just how tricky are the spam bots these days? Would having the image name be the ISBN # be too obvious?

Here is a sample cover:

_{(source: mfrl.org)}

回答1:

You need to make it difficult for an OCR system to read the text. Otherwise the spam bot will easily get through your captcha, without any customisation from a human spammer.

That's why you see funny XORing, noise and distortion on most captchas these days.

As a matter of principle, it makes sense to NOT base the image name on something that can be looked up, although in the case of a local library, chances are low that any spammers will be writing custom scripts to defeat your captcha...