Image recognition to defeat automation.

[Lunawzer]

Hello, first time poster but long time reader here 🙂

I was trying to find out more about the image recognition system used to prevent automation. I don't know the proper term for it and was hoping someone could point me in the right direction.

When you login or register in many online websites, you're presented with a bit of garbled text (e.g. if you try to register for a yahoo mail account) that you're supposed to type in, this stops automated registrations, etc.

I was hoping someone could enlighten me about this.

Thanks.

 

[TuxDave]

Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

 

[Armitage]

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

 

[TuxDave]

Originally posted by: Armitage

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

Haha... human distributed computing at its best. But I was thinking that since some websites use non-garbled text images to prevent automation, and combined with the fact that OCR (optical character recognition) is fairly mature, it shouldn't be hard to make a program that'll read the image and return the text.

 

[jagec]

Originally posted by: TuxDave

Originally posted by: Armitage

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

Haha... human distributed computing at its best. But I was thinking that since some websites use non-garbled text images to prevent automation, and combined with the fact that OCR (optical character recognition) is fairly mature, it shouldn't be hard to make a program that'll read the image and return the text.

Seriously, it would seem that with just a couple image manipulations, a standard OCR program would be able to read those pictures no problem.

 

[BEL6772]

Are you looking for input on how it's done or why it's done?

 

[Armitage]

Originally posted by: jagec

Originally posted by: TuxDave

Originally posted by: Armitage

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

Haha... human distributed computing at its best. But I was thinking that since some websites use non-garbled text images to prevent automation, and combined with the fact that OCR (optical character recognition) is fairly mature, it shouldn't be hard to make a program that'll read the image and return the text.

Seriously, it would seem that with just a couple image manipulations, a standard OCR program would be able to read those pictures no problem.

I doubt it ... the OCR I've used in the past had enough trouble with normal fonts. And some of the anti-scripting images I've seen were hard even for a human ... or maybe it was just me 😀

But seriously, highly stylized letters, all warped & twisted, varying contrast, extraneous lines & marks through them. Maybe OCR has gotten alot better since I saw it last, but I'd be suprised if it got anywhere on some of these.

 

[icarus4586]

Basically, the server has a random text generator and a program that converts that into an image, then garbles it somewhat. No image recongnition or anything.

 

[Zero Plasma]

bump

 

[TuxDave]

Originally posted by: Armitage

Originally posted by: jagec

Originally posted by: TuxDave

Originally posted by: Armitage

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

Haha... human distributed computing at its best. But I was thinking that since some websites use non-garbled text images to prevent automation, and combined with the fact that OCR (optical character recognition) is fairly mature, it shouldn't be hard to make a program that'll read the image and return the text.

Seriously, it would seem that with just a couple image manipulations, a standard OCR program would be able to read those pictures no problem.

I doubt it ... the OCR I've used in the past had enough trouble with normal fonts. And some of the anti-scripting images I've seen were hard even for a human ... or maybe it was just me 😀

But seriously, highly stylized letters, all warped & twisted, varying contrast, extraneous lines & marks through them. Maybe OCR has gotten alot better since I saw it last, but I'd be suprised if it got anywhere on some of these.

www.urbandictionary.com uses un-garbled image recognition. You can see it by going to arandom word and select 'remove' and scroll way down. I have no doubt that OCR can break past that.

 

[Armitage]

Originally posted by: TuxDave

Originally posted by: Armitage

Originally posted by: jagec

Originally posted by: TuxDave

Originally posted by: Armitage

Originally posted by: TuxDave
Wha?

I don't think there is an image recognition system put into place. I thought there was just a bank of preset garbled pictures and they select a random one with its associated word to the webpage. When you login and type in what you see, they just make sure it matches the associated word.

However, if you are good with image processing I think you can defeat this system.

What I've heard of being done to hack these systems was using your script to grab the image, and then posting it as your own "image recognition" test for some high volume site (often pron), then taking the value recovered from there back to the original site.

Haha... human distributed computing at its best. But I was thinking that since some websites use non-garbled text images to prevent automation, and combined with the fact that OCR (optical character recognition) is fairly mature, it shouldn't be hard to make a program that'll read the image and return the text.

Seriously, it would seem that with just a couple image manipulations, a standard OCR program would be able to read those pictures no problem.

I doubt it ... the OCR I've used in the past had enough trouble with normal fonts. And some of the anti-scripting images I've seen were hard even for a human ... or maybe it was just me 😀

But seriously, highly stylized letters, all warped & twisted, varying contrast, extraneous lines & marks through them. Maybe OCR has gotten alot better since I saw it last, but I'd be suprised if it got anywhere on some of these.

www.urbandictionary.com uses un-garbled image recognition. You can see it by going to arandom word and select 'remove' and scroll way down. I have no doubt that OCR can break past that.

Yep, that looks pretty easy. Take a look at the one you get to create a yahoo mail account. I've seen tougher ones also, but I don't recall where.

 

[Lunawzer]

Thanks for the input everyone. I've been doing some research for a class project that involves this. One of the interesting links i've come across is this:
http://www.captcha.net/
More comments and feedback appreciated.

 

[TuxDave]

Originally posted by: Lunawzer
Thanks for the input everyone. I've been doing some research for a class project that involves this. One of the interesting links i've come across is this:
http://www.captcha.net/
More comments and feedback appreciated.

Thanks for the link. I know people who are interested in this type of stuff.

 

[TuxDave]

O_O Someone figured out a clever way to do it. I just got an email in my yahoo account and they wanted to 'verify that my account was registered to real people' so then they gave a link that showed the scrambled text that yahoo gives and a textbox to submit what you see. When you submit it, it goes to a spam company and I'm betting that they use that to automate account creations.