China Captchas

[imron]

Split from discussion here..

 

It's deserving enough for it's own thread because I think many people ignore the baidu account verification thread and might otherwise miss it.

 

--

 

I must say, I'm pretty impressed with the Baidu captcha verification thing they've got going on.

 

They have multiple look-alike characters and sound alike characters for confusing image recognition and other automated systems.  The correct characters are also using different fonts in the question and answer and have different types of distortion applied.

 

It's really quite clever.

[889]

As I pointed out here before, if you want a challenging captcha challenge try www.12306.cn

And what's not impressive about baidu is the security failure that makes the links posted here work.

[imron]

I'm not so sure that's a security failure.  There aren't really a lot of measure you can take to prevent that, that wouldn't also stop a large amount of legitimate requests.  It's more to ensure there is a human involved somwhere and to prevent automated systems from mass registration and I think they've been very clever in designing something that is not too tricky for humans, but still quite difficult for machines.

[imron]

if you want a challenging captcha challlenge try www.12306.cn

"How hard could it be?", I foolishly thought.  Then.... Holy crap!  That's impossible even for humans to solve most of the time (I failed several attempts even when I was seriously trying).

 

Here's a random sample.  I challenge anyone to successfully point out all peanuts and electrical generators in the following image.  Note: this is the original image size and I haven't done anything to distort or shrink it.  Also note: you have to get all of them, and can't incorrectly miss one or accidentally include one that is not one of the requested items.

 

[Bigdumogre]

Where are the peanuts lol

[William2015]

Well, you need a magnifying glass to see them clearly.

[vellocet]

The official China Rail website at http://www.12306.cn/ has the most difficult captchas I have ever seen.  They defeated a native speaker again and again, for over 30 minutes.  Even the times I could read the characters, and clicked the pictures I thought were correct, it was still wrong.  Try to sign up for a new account and see if you can pass.  Bonus: it pops up again when you log in every time. 

[889]

There's already a thread on the China Rail captcha.

http://www.chinese-forums.com/index.php?/topic/49411-china-captchas/

[imron]

merged

[somethingfunny]

Imron, I see 3 electrical generators and what looks like one bowl of 老醋花生米, does that count?

 

I watched a TED talk once about how they use captcha as a crowdsource method to digitise scanned volumes of books.  Basically, they have text recognition software that will recognise 90% (this is a guess) of text.  The remaining text they put into captchas.  Then when you log in somewhere you get two captchas, the first is text it has recognised correctly to verify that you are human and the second is text it couldn't recognise.  It does the same for ~20 people and compares their guesses for the unrecognised text.  If enough of them typed in the same thing then you've managed to digitise the text.  Congratulations.

 

Here is a link to the talk which is definitely worth a watch.  The guy is pretty cool and in the second half he talks about how his research group is using the same principle to translate the internet while helping people learn a foreign language.  I don't watch a lot of TED talks, some of them seem to get a bit too far away from real research and veer into motivational seminar territory, but this one was really good and gave me a real 'humans are f**king awesome' feeling.

 

I don't know a lot about captchas (thankfully) but it seems like they need to get China on board to this kind of thing.

[imron]

Relevant xkcd