On Mon, 2007-06-11 at 20:47 +0200, Tijnema wrote: > On 6/11/07, Robert Cummings <robert@xxxxxxxxxxxxx> wrote: > > On Mon, 2007-06-11 at 14:29 -0400, Daniel Brown wrote: > > > > > To be a bit easier, I whipped up a quick example on the web. It's > > > > > just static images, not a working system, but you'll see what we're > > > > > getting at here: > > > > > http://pilotpig.net/captcha-example.php > > > > > > > > > > > > > You could even use a color check here to see which color matches the best ;) > > > > > > > > > > Ah, but that validates my exact point --- the system will see the > > > overlay over the legitimate cigarette image as being image-synonymous > > > with the snake as an option by color pattern, while the parent image > > > matches best with the first child option image in shape. > > > > Not as easy as Tijnema thinks... all the icons are in a single image so > > first he needs to find the icon boundaries to extract them to perform > > colour analysis. And that can be more or less hard depending on how the > > icons are merged. For instance using PNG images with alpha transparency > > so that an overlay and merge looks right would make edge detection of > > the icon difficult ths making colour analysis difficult. > > Sure, but what if I convert the image first to JPEG or GIF? GIF would > be the easiest option I think, because if I convert both then I could > easily count the color of each pixel and you can call the job done. :) No, I'm talking multiple PNG images. Each icon is a PNG image with appropriate boundary alpha transparency that hugs the shape. These are added to the primary display CAPTCHA thus the edges blend nicely with the master background for the given CAPTCHA in the RESULTING image. Thus you can convert to GIF all you want... it won't help you with edge detection / colour analysis. > > Also, the > > colour analysis only works in the case where you're presented with an > > image and asked to pick the same image form the set. It doesn't work in > > the semantic example where you are asked "which of the following doesn't > > belong?" :) > > Of course, it was just an easy example, as there could be images > presented with exactly the same color. Different CAPTCHA program needs > different kind of hack... As already stated this particular method employs multiple methods. It would require multiple approaches and then again would require the cracking programs knows what constitutes what approach :) Imagery is much more difficult to crack than simple text. Take 5 pictures of apples, different apples for fun ;), to each all but one apple add a smaller sub icon a leaf, again to all but one apple add another sub-icon, let's say a worm. Finally, let's add one more sub-icon, a pie (mmmm apple pie ;). Now the challenge is: What icons are present with the apple that doesn't have a leaf. Good luck :) Pictures within pictures are very complex and then requiring semantic understanding of those pictures and the question asked is even harder. Cheers, Rob. -- .------------------------------------------------------------. | InterJinn Application Framework - http://www.interjinn.com | :------------------------------------------------------------: | An application and templating framework for PHP. Boasting | | a powerful, scalable system for accessing system services | | such as forms, properties, sessions, and caches. InterJinn | | also provides an extremely flexible architecture for | | creating re-usable components quickly and easily. | `------------------------------------------------------------' -- PHP General Mailing List (http://www.php.net/) To unsubscribe, visit: http://www.php.net/unsub.php
