helraizer Posted January 30, 2008 Share Posted January 30, 2008 Hi folks, I was just wondering, some captcha images can be read by OCR image analysers because they read the characters from the image and guess due to the width and shape etc.. I've noticed that most CAPTCHA images are use a single font, would it be harder for the bot/analyser to read the characters if ever character in the string was a different font (including size) or would it make no difference? Sam Quote Link to comment Share on other sites More sharing options...
The Little Guy Posted January 30, 2008 Share Posted January 30, 2008 It depends on the bot/analyzer. Some might search many different types of fonts for each letter. some may not. So far I have never had any spam with a CAPTCHA though. A new type of CAPTCHA I came across was kind of cool... it asked a question, such as: "A dog has __ legs" then there is a text box that you fill in the blank for the answer. Quote Link to comment Share on other sites More sharing options...
PFMaBiSmAd Posted January 30, 2008 Share Posted January 30, 2008 OCR basically identifies end points of lines or points along a curve and then connects the dots. It does not really matter if different fonts or different size characters are used in one image. Doing so might make the image look more confusing to a human, but to a computer, no. Quote Link to comment Share on other sites More sharing options...
The Little Guy Posted January 30, 2008 Share Posted January 30, 2008 Using weird fonts may help in confusing a computer, or using crazy transparent foregrounds in front of the text. This makes it harder to read for a human, but can be readable if the foreground is carefully created. Here is an example CAPTCHA that I created: http://phpsnips.com/examples/CAPTCHA/form2.php If you want the source code: http://phpsnips.com/snippet.php?id=43 Quote Link to comment Share on other sites More sharing options...
cooldude832 Posted January 30, 2008 Share Posted January 30, 2008 what I have come to realize with captcha is there is 3 levels of them Level 1 is a super basic background + colored straight roman time easy to read, and force you to read it , a very simple OCR can read it, but at least your users aren't mad they can't read it Level 2 is a broad range between insanely hard and level 1 slows down the OCR but still crackable Level 3 is what google uses and I hate it because I can never get them right Personally if you have additional filters on you are fine with level 1 to a low level 2 the point is that it needs to be typed and that slows down 99% of the user base Quote Link to comment Share on other sites More sharing options...
helraizer Posted January 31, 2008 Author Share Posted January 31, 2008 What I did to make my captcha was use an obscure font with 2 codes written on it (but in the background, unseeable by humans about 17 other codes) then have the random code as a string and split the string into 1 character chunks, also have a random number. Depending on what the random number is, let's say 1 - 3 then the top code in the image is the one to use and 4 - 6 (not the exact numbers) then the bottom code is to use. Also (now the reason for the 1 character chunks) if the number is 1 then the user is requested to enter characters 1,3,6 from the top image if the the number is 4 the characters 6,2,5 from the lower code (in order (not exact order but it's the idea that counts)) etc... (It's easier to use than it is to explain ) Would that help to stop a OCR? This one (as described): or this one: again the user has to put in a certain order, which means that the OCR bot can read it but then has to guess which order to put them in. There is around a 1 in 216 chance for the bot to get it.. if I ask for 4 characters then that increases further to 1 in 576 change then if my maths is right, if 5 characters are needed there is a 1 in 1536 chance. Therefore the bot isn't all that likely to guess easily. If the same goes for the top image, those chances double so for 3 characters there is a 1 in 432 chance, for 4 = 1 in 1152 and for 5 = 1 in 3,072 chance since there a two codes. So would that be hard for a bot since once it's read it it's all guess work? Sam Quote Link to comment Share on other sites More sharing options...
helraizer Posted January 31, 2008 Author Share Posted January 31, 2008 Made an error in that last post but less than a minute later I can't edit the post ??? Ah well. The top image is this Quote Link to comment Share on other sites More sharing options...
cooldude832 Posted January 31, 2008 Share Posted January 31, 2008 your siggy almost got me, Iwas like wait you only listed two and you said oooooh (10 = 2 in binary ) Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.