site: reCPTCHA

[AF]

http://recaptcha.net/

A new form of captcha (currently doing the rounds on BBC news). It takes text illegible to OCR from book digitizing projects and uses them as captchas, providing a known and an unknown. As a result the captcha can be guaranteed as illegible to bots, and users actively contribute to book digitization efforts by deciphering what OCR fails to read.

The project has provided patches for common projects including phpbb and media wiki.

[clericvash]

Thanks for the link, will be using that on my website.

[Tim Blokdijk]

We already use it for the wiki.
http://spring.clan-sy.com/w/index.php?t ... ype=signup

[Tim Blokdijk]

Now that you bring it up, is that recaptcha ocr improvement stuff publicly available? Not that I would like to remove it if it's commercial, if it work then it works after all. I just like to know if I get better ocr software on my Linux system if we use this.

[AF]

Actually according to their site its best we stick with them as there's a large stockpile of illegible text and its a central system.

Should reCAPTCHA be thwarted by spammers they can fix the problem and we don't have to do anything. The centralized nature also allows them to track global spam bot epidemics and react faster. Its a reliable outsourcing of the spam bot problem in order to serve a dual purpose.

[SinbadEV]

I don't think it's improving OCR, we are just telling the computer what that word means, thereby distributing the load of transcribing texts across the internet. Meanwhile if the spammers make an OCR awesome enough to crack it they could probably sell it for more then they are making spamming.