Since it is widely know that the captcha implementations used by Yahoo, Microsoft, and Google can be broken by computers with a fairly high success rate (around 20-30%), I have been throwing around ideas in my head about a new captcha implementation. However, I don't have a workable or complete idea. I think I need to browse some math sites until something dawns on me.
Basically, I want to do away with image/text recognition and use pattern recognition instead since humans are much better than computers at pattern recognition. The pattern would be generated as an image by the host server on the fly to allow for an astronomically high number of possibilities. There would only be one pattern generated for each captcha request but the user is required to click on a series of points on the image that correspond to a pattern. The host server holds onto a set of coordinates that correspond to a correct sequence of mouse clicks and gives validation only if the user clicks on points close to this key pattern. The client side captcha is essentially a simple javascript that returns the mouse coordinates when the user clicks on the image of the pattern. This multiple mouse click requirement coupled with an image about 500x500 should provide a fairly high level of bot protection while keeping it simple to use.
The problem is that I need a method to generate patterns that can't be easily recognized by computers but is recognizable by humans.
Perhaps we should have some sort of captcha contest on DonationCoder to see who can come up with the best implementation? Also, let's get a list of new/innovative captcha implementations going so we can see what has been done.
