Here's another:
Generate a complex image of some realistic looking scene, with some noise
perhaps to make it hard to do machine vision on, then ask them to locate
one or more objects in the scene. The coordinates are the result. If you
gave them a reasonably large image then you can get a lot of information
this way, specifically:
n*(log2 w*h)
bits of it.
On Wed, May 02, 2001 at 12:19:48AM -0500, Scott G. Miller
wrote: > >
> > Now you just need to suggest a puzzle which is machine generatable, not
> > machine solvable, and not too obnoxious.
>
> Heres one idea, but I dont think it generates enough information for a
> key. I define "enough information" to be greater than 2^32 choices for
> someone to do a brute force insertion to cover all the possibilities
> mechanically.
>
> Easy: Generate an image that contains a polygon formed from sparse,
> non-connected dots more densly packed where the letter is. Make it
> significantly random. Machine vision programs will suck at this, neural
> nets can do it but they have to be fairly large and will be slow.
> Then present a 4 x 4 tile of line drawings. Ask the user to pick which
> one looks like the dotgram.
>
>
> The key is to pack more information into the presented image, and allow
> for many choices. Ever used Qbist in the GIMP? A good system would
> present a fairly complicated dotgram that required several directional
> steps in a Qbist like interface. Each choice would bring the image closer
> to the real one. This might be too annoying however.
>
> I'll think about some other ones.
>
--
rest greg trim coolly
PGP signature
