No, he just hard-coded the number shapes. As long as the numbers used the same font as the example (and don't overlap), it should work just fine.
OCR would probably be more robust in general (for varying fonts and number shapes)... but it's simply absurd to call G-WAN's scheme a better captcha. More obscure and less targeted? Perhaps.
I thought about using OCR - just to over-engineer it. But I wanted to show how the characters are perfectly aligned and how clear the font is. I would like to understand, what he thought, why this Captcha is so special.
Re:G-WAN Captcha Decode
by merlincorey on Thursday June 14, 2012 @06:12PM
(#4113795)
I need two more points to make up for the haters.
Come find us?
Both you and the OP have misunderstood the purpose of the CAPTCHA example. See my post[1] for an explanation.
Why don't you identify yourself as the author?
I am not the author. For "proof", see my many conversations on the G-WAN forum.[1]
[1]http://forum.gwan.com/index.php?p=/profile/discussions/1053/...
I don't think he is. I read the G-WAN authors replies. The author doesn't have as good grasp of English and he sounds way crazier.
Re:G-WAN Captcha Decode
by merlincorey on Thursday June 14, 2012 @06:05PM
(#4113769)
I haven't ran the code, but I did completely read it.
It appears to me that he maps the character set GWAN uses for the captcha system, and as such, should work for any image generated by GWAN using the same character set as it simply identifies pixels matching the character set.
In any case, it seems trivially easy to break. Just capture the image. Read the background color value. Generate the image (with the background color) in ImageMagick and run through your OCR of choice. Obviously, that's not the fastest way to do it if you're trying to do thousands of attempts at once, but it's the least brainpower-involved.