reCAPTCHA - Did You Know?

howyoudoin · 11-05-2011, 06:05 AM

How many times have you been waiting to download a file, and come across a 'toll booth' in the form of a reCAPTCHA window?

What you probably didn't realise, as you keyed in the letters with one hand, was that you were actually a tiny component in the grand design to OCR old Texts for archival purposes.

Every now and then, automated OCR scripts come across words on pages that are just indecipherable for various reasons. These words end up in those tiny reCAPTCHA windows, waiting for the ultimate OCR machine (you!) to transcribe them! The reason file hosting sites, or sign-up pages are confident that spambots can't get through that barrier is because these are words that have stumped automated text recognition programs to begin with!

From wikipedia:

reCAPTCHA is a system originally developed at Carnegie Mellon University that uses CAPTCHA to help digitize the text of books while protecting websites from bots attempting to access restricted areas.[1] On September 16, 2009, Google acquired reCAPTCHA.[2] reCAPTCHA is currently digitizing the archives of The New York Times and books from Google Books.[3] Twenty years of The New York Times have been digitized and the project planned to have completed the remaining years by the end of 2010.[4]

reCAPTCHA supplies subscribing websites with images of words that optical character recognition (OCR) software has been unable to read. The subscribing websites (whose purposes are generally unrelated to the book digitization project) present these images for humans to decipher as CAPTCHA words, as part of their normal validation procedures. They then return the results to the reCAPTCHA service, which sends the results to the digitization projects.

The system is reported to display over 200 million CAPTCHAs every day,[5] and among its subscribers are such popular sites as Facebook, TicketMaster, Twitter, 4chan, CNN.com, and StumbleUpon.[6] Craigslist began using reCAPTCHA in June 2008.[7] The U.S. National Telecommunications and Information Administration also used reCAPTCHA for its digital TV converter box coupon program website as part of the US DTV transition.[8]

NOW YOU KNOW!

Asawi · 11-05-2011, 06:35 AM

I'm not surprised. Some of them are ridiculously hard to decipher!

I have to do it every time I check out an eBook from the Swedish eLibrary and I do that a lot!). I haven't had any problems lately, but earlier this year I had to refresh multiple times to be able to make out the letters (or strange symbols that I had no idea how to recreate on my keyboard)!

SmokeAndMirrors · 11-05-2011, 06:49 AM

I had heard this before, but didn't know whether it was true. Neat!

@Asawi - It's certainly possible that some of the characters aren't represented anymore in modern versions of various languages. The English language lost quite a few characters during the periods of middle and early modern English, because they apparently didn't care enough to make their own typefaces for print, and imported them from other European countries instead. Problem is, a lot of the languages in those countries didn't use all of the letters that English did, and therefore their typefaces didn't include them.

In some cases this culling of the English alphabet was for the best - wynn, for example, was a completely redundant letter. But in other cases, like with thorn, useful and logical letters were lost. Some of these letters were used up until just a couple hundred years ago (often substituted with other Latin letters in print).

So, it is certainly possible that you've run across letters that simply don't exist anymore. And there is indeed no way for you to represent them on your keyboard - at least not without special characters.

Richey79 · 11-05-2011, 07:59 AM

Now that is interesting.

4Chan users helping to OCR venerable archived texts as they upload vile images... a pretty amusing irony.

I do seem to get truly undecipherable stuff pretty often: upside down Hebrew as an example. I wonder what processes they use to work out which responses are valid and can be relied on.

Hamlet53 · 11-05-2011, 08:00 AM

That is interesting. So all guesses are logged, even those guesses that at the time do not match the answer? Continuing update of the answer to match the most frequent guess?

Well in any case this information does not alter the fact that I find those reCAPTCHA barriers one of the most annoying things to encounter on the web.

PeterT · 11-05-2011, 08:41 AM

Maybe I am confused... How on earth can an answer to a RECAPTCHA help OCR? In order for you tp get through the check, you have to type the letters that correspond to the image.

In order for RECAPTCHA to know you are correct, it has to compare your result to it's known result.

So to my eyes that says that the image has already been decoded at least once.

howyoudoin · 11-05-2011, 08:46 AM

Quote:

Originally Posted by PeterT

Maybe I am confused... How on earth can an answer to a RECAPTCHA help OCR? In order for you tp get through the check, you have to type the letters that correspond to the image.

In order for RECAPTCHA to know you are correct, it has to compare your result to it's known result.

So to my eyes that says that the image has already been decoded at least once.

Like Hamlet53 mentioned, it probably aggregates a lot of user entries for each word from the reCAPTCHA, and picks the most frequent one as the 'correct' answer to be sent back to the archival system. As far as the barrier purpose goes, it probably already has an approximate answer (but obviously not an entirely satisfactory one) provided by the initial OCR software, and as long as it sees that your answer is close enough to the answer it initially had, you'll get through. I think that's the case because there have been instances where I've not supplied an exact answer, but I've still been let through.

JoeD · 11-05-2011, 09:52 AM

You have to enter two words, one a known word and the other a unknown ocr based one. You can type anything you want for one of the two so long as the known word is entered correctly.

captchas are annoying yes, but they do limit spam a little at least which would be even more annoying.

pidgeon92 · 11-05-2011, 02:21 PM

I read this years ago. I think about it every time I do one of them.

RainingLemur · 11-05-2011, 02:47 PM

I think it's pretty cool. It'd be nice to see the fruits of the "labor" though. Maybe some kind of report that shows what works have been completed with the use of this method.

Rob Lister · 11-05-2011, 03:38 PM

Captcha is what finally convinced me that I'm probably not human.

carpetmojo · 11-05-2011, 07:17 PM

I kept trying new glasses 'till I worked it wasn't necessarily my fault I didn't know what I was looking at.................

p'raps it's Esperanto ?

taustin · 11-05-2011, 11:59 PM

Quote:

Originally Posted by PeterT

Maybe I am confused... How on earth can an answer to a RECAPTCHA help OCR? In order for you tp get through the check, you have to type the letters that correspond to the image.

In order for RECAPTCHA to know you are correct, it has to compare your result to it's known result.

So to my eyes that says that the image has already been decoded at least once.

That's why there are two nowadays. One is a true OCR item, the other is made to order, so that it does, indeed, know the answer. Usually, the less legible one is the known one; the more legible is usually the real OCR one.

What's less well known is that some spam operations operate "free" porn sites for the specific purpose of getting people to decipher the OCR stuff, which they then use to sign up for free email accounts (having gotten the image from the sign up page for the email). This allows them to completely automate the process of getting free email accounts to spam from. The sad thing is, these are the less slimy spammers.

afa · 11-06-2011, 11:07 AM

Very interesting.

Like many others, I find the reCAPTCHA extremely annoying. Quite fascinating to find out what it's real purpose is.

Richey79 · 11-07-2011, 06:50 AM

Quote:

Originally Posted by JoeD

You have to enter two words, one a known word and the other a unknown ocr based one. You can type anything you want for one of the two so long as the known word is entered correctly.

Only problem is, a lot of people know this, so just type 'wot' or some gibberish for every one of the non-recognised answers to speed things up.

11-05-2011, 06:05 AM	#1
howyoudoin how YOU doin? Posts: 1,100 Karma: 7371047 Join Date: Feb 2009 Location: India Device: Kindle Keyboard, iPad Pro 10.5”, Kobo Aura H2O, Kobo Libra 2	reCAPTCHA - Did You Know? How many times have you been waiting to download a file, and come across a 'toll booth' in the form of a reCAPTCHA window? What you probably didn't realise, as you keyed in the letters with one hand, was that you were actually a tiny component in the grand design to OCR old Texts for archival purposes. Every now and then, automated OCR scripts come across words on pages that are just indecipherable for various reasons. These words end up in those tiny reCAPTCHA windows, waiting for the ultimate OCR machine (you!) to transcribe them! The reason file hosting sites, or sign-up pages are confident that spambots can't get through that barrier is because these are words that have stumped automated text recognition programs to begin with! From wikipedia: reCAPTCHA is a system originally developed at Carnegie Mellon University that uses CAPTCHA to help digitize the text of books while protecting websites from bots attempting to access restricted areas.[1] On September 16, 2009, Google acquired reCAPTCHA.[2] reCAPTCHA is currently digitizing the archives of The New York Times and books from Google Books.[3] Twenty years of The New York Times have been digitized and the project planned to have completed the remaining years by the end of 2010.[4] reCAPTCHA supplies subscribing websites with images of words that optical character recognition (OCR) software has been unable to read. The subscribing websites (whose purposes are generally unrelated to the book digitization project) present these images for humans to decipher as CAPTCHA words, as part of their normal validation procedures. They then return the results to the reCAPTCHA service, which sends the results to the digitization projects. The system is reported to display over 200 million CAPTCHAs every day,[5] and among its subscribers are such popular sites as Facebook, TicketMaster, Twitter, 4chan, CNN.com, and StumbleUpon.[6] Craigslist began using reCAPTCHA in June 2008.[7] The U.S. National Telecommunications and Information Administration also used reCAPTCHA for its digital TV converter box coupon program website as part of the US DTV transition.[8] NOW YOU KNOW! Last edited by howyoudoin; 11-12-2011 at 12:53 PM.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Oh no, watch out for P word in your e-texts via reCaptcha	joedevon	News	13	07-24-2009 06:12 PM

11-05-2011, 06:35 AM	#2
Asawi Gadgetoholic Posts: 1,462 Karma: 2731820 Join Date: Feb 2011 Location: Sweden Device: Kobo Libra2, Tolino Vision 6	I'm not surprised. Some of them are ridiculously hard to decipher! I have to do it every time I check out an eBook from the Swedish eLibrary and I do that a lot!). I haven't had any problems lately, but earlier this year I had to refresh multiple times to be able to make out the letters (or strange symbols that I had no idea how to recreate on my keyboard)!

11-05-2011, 06:49 AM	#3
SmokeAndMirrors Addict Posts: 280 Karma: 2064388 Join Date: Aug 2011 Location: MN, US Device: Kobo Touch, Asus Eee Pad Slider	I had heard this before, but didn't know whether it was true. Neat! @Asawi - It's certainly possible that some of the characters aren't represented anymore in modern versions of various languages. The English language lost quite a few characters during the periods of middle and early modern English, because they apparently didn't care enough to make their own typefaces for print, and imported them from other European countries instead. Problem is, a lot of the languages in those countries didn't use all of the letters that English did, and therefore their typefaces didn't include them. In some cases this culling of the English alphabet was for the best - wynn, for example, was a completely redundant letter. But in other cases, like with thorn, useful and logical letters were lost. Some of these letters were used up until just a couple hundred years ago (often substituted with other Latin letters in print). So, it is certainly possible that you've run across letters that simply don't exist anymore. And there is indeed no way for you to represent them on your keyboard - at least not without special characters.

11-05-2011, 07:59 AM	#4
Richey79 mrkrgnao Posts: 241 Karma: 237248 Join Date: May 2010 Device: PRS650, K3 Wireless, Galaxy S3, iPad 3.	Now that is interesting. 4Chan users helping to OCR venerable archived texts as they upload vile images... a pretty amusing irony. I do seem to get truly undecipherable stuff pretty often: upside down Hebrew as an example. I wonder what processes they use to work out which responses are valid and can be relied on.

11-05-2011, 08:00 AM	#5
Hamlet53 Nameless Being	That is interesting. So all guesses are logged, even those guesses that at the time do not match the answer? Continuing update of the answer to match the most frequent guess? Well in any case this information does not alter the fact that I find those reCAPTCHA barriers one of the most annoying things to encounter on the web.

11-05-2011, 08:41 AM	#6
PeterT Grand Sorcerer Posts: 12,260 Karma: 74007256 Join Date: Nov 2007 Location: Toronto Device: Nexus 7, Clara, Touch, Tolino EPOS	Maybe I am confused... How on earth can an answer to a RECAPTCHA help OCR? In order for you tp get through the check, you have to type the letters that correspond to the image. In order for RECAPTCHA to know you are correct, it has to compare your result to it's known result. So to my eyes that says that the image has already been decoded at least once.

11-05-2011, 09:52 AM	#8
JoeD Guru Posts: 895 Karma: 4383958 Join Date: Nov 2007 Device: na	You have to enter two words, one a known word and the other a unknown ocr based one. You can type anything you want for one of the two so long as the known word is entered correctly. captchas are annoying yes, but they do limit spam a little at least which would be even more annoying.

11-05-2011, 02:21 PM	#9
pidgeon92 Wizard Posts: 3,144 Karma: 8426142 Join Date: Jun 2008 Location: Chicago, IL Device: Kindle PW2, Kindle Voyage, Kindle DXG, Boox M90, Kobo Aura HD	I read this years ago. I think about it every time I do one of them.

11-05-2011, 02:47 PM	#10
RainingLemur Spork Connoisseur Posts: 2,355 Karma: 16780603 Join Date: Mar 2011 Device: Nook Color	I think it's pretty cool. It'd be nice to see the fruits of the "labor" though. Maybe some kind of report that shows what works have been completed with the use of this method.

11-05-2011, 03:38 PM	#11
Rob Lister Fanatic Posts: 532 Karma: 3293888 Join Date: Oct 2011 Location: Virginia Device: Nook Simple Touch	Captcha is what finally convinced me that I'm probably not human.

11-05-2011, 07:17 PM	#12
carpetmojo Wizard Posts: 3,117 Karma: 9269999 Join Date: Feb 2011 Location: UK Device: Sony- T3, PRS650, 350, T1/2/3, Paperwhite, Fire 8.9,Samsung Tab S 10.5	I kept trying new glasses 'till I worked it wasn't necessarily my fault I didn't know what I was looking at................. p'raps it's Esperanto ?

Advert

Advert

11-06-2011, 11:07 AM	#14
afa The Forgotten Posts: 1,136 Karma: 4689999 Join Date: Apr 2010 Location: Dubai Device: Kindle Paperwhite; Nook HD; Sony Xperia Z3 Compact	Very interesting. Like many others, I find the reCAPTCHA extremely annoying. Quite fascinating to find out what it's real purpose is.