> To us it's easy to pick out a dark square in the light or a
> white square in the shadow, but a computer would need some
> complex processing rather than just comparing pixel values.
> Looking for pixel values of (120,120,120) isn't going to
> find and distinguish between light and dark checkers.
> David Coombes

David :

  Thanks you most sincerely for that explanation (below) , and
please forgive my top posting an edited version of your reply ,
 ... but I think the whole point of that link was that we _can-
not_ easily tell the difference at all ( I aplogise for what
an awefull mess the MicroSlop mailer does to your replies) .

  Anyway , Timo's initial link that Yogi resent to the list :
http://images.google.de/images?q=checker+illusion&hl=de&btnG=Bilder-Suche
shows just how bad our perceptions can be in extreme situations .

Garry Curtis
http://www.niagara.com/~studio


> Not related to 3D work but image recognition, so off topic. Present image
> recognition tends to work by finding similar shapes or following edge
> contrast. Stick a camera on a computer and it'll capture a colour value
> for
> each pixel. From that it can try to guess what it's looking at by
> programming shape recognition etc. In the example picture it'll see the
> same
> level of grey (120,120,120) for the dark squares in the light and white
> squares in the shadow, and not be able to recognise there's a difference.
> A
> human viewer doesn't see each pixel of the image, but an overall image.
> The
> onlooker sees the green cylinder, the apparent shadow, the pattern of
> squares, and concludes it's a 3D scene with checkerboard in shadow, so
> adjusts the perception of the colours accordingly. To us it's easy to pick
> out a dark square in the light or a white square in the shadow, but a
> computer would need some complex processing rather than just comparing
> pixel
> values. Looking for pixel values of (120,120,120) isn't going to find and
> distinguish between light and dark checkers.
> David Coombes
> [EMAIL PROTECTED]
















Reply via email to