Well, I know lots of you guys must've cracked, or tried to crack a CAPTCHA or two.. or more 

So I was wondering if you guys used neural

net

 s to do it or not, and if you did, what did you use as input? 

Just delete this post if you think it's passing some invisible boundries .

Well, no real boundaries here (except those listed in forum rules...) but, that info would certainly not be posted or discussed here because of the proprietary nature...

I'm fairly certain that anyone here, if they actually had such experience would not share it in such an open forum.

I'm gonna speculate that there's other, better suited forums for that discussion...

Use this to clean the captcha
http://www.intel.com/technology/computing/opencv/
(make it black and white and clean out the noise)
and then isolate the individual letters.

once you have the letters isolated into like 16x16 pixel blocks feed them into an opensource neural

net

  suitable for OCR.

The vision library has a

python

  binding (probably only works on

linux

  

i never pursued this past the filtering the captcha, converting to black and white and isolating the letter blocks.
(search on .edu sites for how to do this, look for computer vision).

There are better ways to break captchas for me and too much investment of time for me

Oh well crap, I forgot about Nop... 

LOL perkiset..

nop : I used some of my own self created algos (lol ) to clear the noise and stuff, and put the characters into 30x30 blocks and then just tried to matches those with a db of characters, where match is see which has more pixels in common..

needless to say.. i can only go so far with my way .. + it gets kinda slow with sometimes up to 1 second processing.. (all in

python

 
thanks for the info!

@Dbyt3r wow . impressive.
just using the opencv functions i filtered the captcha down.
then using edge detection i made it black & white
you should look for

tutor

 ials on the

net

  for opencv and pattern recognition etc.

If you are breaking like a million captchas probably worth while.
or if you are making something like xrumer etc.

I did not see financial viability in breaking capthca

Human neural

net

 s work great imho

I use captcha cracking in my social

net

 work "promotion" apps .

And I have to automate captcha cracking, because my age wouldnt allow me to pay indians to do it .

depending on what you're trying to do and for which site sometimes human semi-automation is the most efficient way.  other methods i've tried are the standard ocr of the image and believe it or not voice recognition because some of the larger technically savy organizations present the vision impaired with an audio recording of the text.

by far though, semi-automation is the widest used imho - captchas are becoming more and more tricky.