[thelist] Security code images

[Matt Warden]

On Tue, 3 Aug 2004 15:12:34 +0100, Richard Davey <rich at launchcode.co.uk> wrote:
> There is another way - instead of scrambling a code and telling the
> user to "type in what you see" - you could show a picture, of say 3
> rabbits and say "how many rabbits do you see?". The user types in 3
> (obviously!).
> 
> It's a simple example, but you can see the subtle difference. No
> script could ever accurately determine what is or isn't a rabbit in a
> photo.

Very nice idea. However, I wonder if it would be trivial to determine
how many times a given pattern displays in an image. What I mean is,
the task you describe does not require that the machine be able to
determine that there are three rabbits, only that there are three of
something. Machines are pretty good at doing this.

I almost think it would be better to show a single rabbit (or
whatever) and ask what it is. But then you'd have to accept 'rabbit',
'bunny', and every other flavor.

Thanks,


-- 

Matt Warden
Berry Neuroscience Lab
Department of Psychology
Miami University



This email proudly and graciously contributes to entropy.