Joshua Shriver <jshriver <at> gmail.com> writes:

> For a while they've had a system where volunteers can "check out" a
> scanned page, check it against OCR output and make corrections.

The problems I can see are:

a) Assuming the Captcha is a piece of original book text warped a little, and
the exact OCR of it is not available, how does one validate the entry supplied
by the reader, since there is no reference to compare it against ? Majority
scoring ? It would be nice to know.

b) Knwin tha awg leeve of (il)litracy and spilling sklls wht xactl do he hop ta
achiev wit dat. Come to think of it, getting the spelling right on a Captcha
test for a p0rn site is a strong incentive for Joe sixpack to polish up his
reading and typing skills (even after absorbing the sixpack).

>From my limited experience with OCR, it is possible to 'guess' the text in
despite of noise and errors by manipulating the picture somewhat randomly and
OCR-ing it repeatedly. This highlights some of the characters sometime. A
program can then reassemble the bits using a best match algorythm. I did not use
this for Captcha but for S/N recognition.

Peter P.


-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist