#1
|
|||
|
|||
Captcha
Hi,
First I'd like to say that JD is quite useful. Second I can summarize the following discussion: http://board.jdownloader.org/showthread.php?t=20134 from my point of view in one sentence: It seems like JD's captcha breaker *NEVER* succeeded in breaking any captcha, for example for sites such as hotlink, fileserver , or filesonic. It seems quite surprising considering this impressive link: http://jdownloader.org/knowledge/wik...orted-captchas and the boasting from the machine learning community that today's captcha is virtually broken. So first I would appreciate if someone would recap what exactly am I missing or what the problem is? Second I'd like to offer a simple way to train a neural network (or something similar) to break such stuff. I think JD should send any successful captcha image and text that was entered by a user to a server, which in turn would save it in a training database. Having such a large DB to learn from, surely the chances of breaking a captch would be higher. Last edited by zoharl; 15.11.2011 at 00:45. |
#2
|
||||
|
||||
Don't think we claimed we could crack all captcha types, for the methods we do not have solutions for the user gets prompted to solve.. math , (unbroken)? circles, colour captchas to text, recaptcha, etc..
Have a look at this thread, http://board.jdownloader.org/showthread.php?t=8216. The captcha type you are trying to solve called recaptcha. They can be supported by external 3rd party applications or porting into APIs, some have created AI based solutions but those are closed. As for solving recaptcha, the team decided to spend time on other projects because fighting google was waste of time. People have discussed the idea of how to break recaptcha and not so long after this google changed recaptcha layouts (this happened a couple times), they where monitoring our forums. Others have since created solve solutions: code based or api(remote solving), these link into our JAC (so no big loss).
__________________
raztoki @ jDownloader reporter/developer http://svn.jdownloader.org/users/170 Don't fight the system, use it to your advantage. :] Last edited by raztoki; 15.11.2011 at 12:12. Reason: refinement. |
#3
|
|||
|
|||
I see, reasonable arguments.
Still although I'm new in this business, I have the feeling that training a neural network with a big enough database could solve any generic captcha. I am wondering if people tried that already, or as your link mentioned they tried a direct method such as separating the letters and make them stand straight. I am willing to give it a try if you can supply me with the data. I think I can organize a server with a database that can receive data. The question is would it be possible to incorporate inside JD a plugin that would send to this server all the successful reCaptcha from every client? |
#4
|
|||
|
|||
In order to do that you need to set up a mechanism and a server like CT and CBH did. You would then have to find jD customers that are willing to participate in your experiment. Also contact SonyTeam because they already have an anti-reCaptcha method for jD.
|
#5
|
|||
|
|||
Yes, I was referring to setting up a server with a database.
I think that customers would be happy to participate in such a project. It cost them nothing, they do nothing, and their automated captcha service is getting better. Moreover I proposed a general solution for all captcha schemes, including new ones to come. Some generic system that would learn from a database. Probably it's no that simple, but I'm willing to give it a try. I'll need to test anti-reCaptcha, although from comments of users on warez-bb.org, it doesn't work. |
#6
|
|||
|
|||
I would test anti-reCaptcha for yourself, because the success rate fluctuates a lot in time. It depends on Gogol's attempts to make them as unreadable as possible, but yet make them still readable by some talented people. I think it won't take long before they give up as nobody will be able to read them and only machines will pass the test.
There have been good algorithms (for instance those of Houck in 2010) but I think only spammers use them to break into systems automatically. It seems that the algorithms were rather simple. Chad Houck and other reCaptcha specialists probably are very rich people now. |
#7
|
||||
|
||||
Couple months back someone did similar AI based system ontop of say antirecaptcha type solver (think they used amazon cloud). They had it open for couple weeks and google was ontop of it quickly (look back in the thread I posted in my original answer, its mentioned within).. Just keep in mind google has invested a lot of money into recaptcha and they'd be over your system like a rash. Not to say it cant be done of course...
__________________
raztoki @ jDownloader reporter/developer http://svn.jdownloader.org/users/170 Don't fight the system, use it to your advantage. :] |
#8
|
|||
|
|||
@zoharl
I think you need to go underground as Gogol has spies everywhere. They spy on your WIFI networks, so don't use it. Also close your curtains, because they'll take pictures. It's like the brave new world. You can't do anything unless the extremely corrupt, power hungry, criminal advertising monopolist Gogol Mafia approves it. |
Thread Tools | |
Display Modes | |
|
|