Task 1) You may be able to do task one using as method similar to the
following:
1) Convert the images to monochrome
2) Increase their contrast without regard to clipping, bring them close
to black & white
3) Compare them, if a (very) close match, you have a hit
4) If no match, invert one of them and compare them again, if a very
close match you have a hit, otherwise you have a miss.
There probably is a bit more to it than that but its a start.
Task 2) That's a tough one...not really sure...something similar to the
MPEG motion estimation algorithm perhaps, extended to account for
rotation of an object? (I'm sure there are better ideas out there.)
Task 3) Not familiar with the CIE algorithim. Is that a heuristics
algorithm (which is something that could work.)
Hope this helps a little.
-Ted
Jmc wrote:
Hi KJ
I need a bit more advanced functionality.
To break it down I would need 3 different recognition functions.
1.
Basically what I would like to do is to have a picture lets say a ball,
the image regognision system should be able to identify an exakt copy
of the picture (OR the same placement of pixels regardless of color).
Imagine that I have a picture of a green ball and some one colors it
yellow, I would still like it to identify it as the "same" picture but
with color variation.
2.
Also I would like to be able to store patterns in a database, lets say
the shape of a balloon, I want my application to be able to identify a
simular shape and then suggest that the picture might illustrate a
baloon.
3. And simularities in colors a red picture should find other red
pictures (but not the exact rgb value but rather a red-ish color) That
part I think I can handle using the CIE algoritm.
The most imortant function is no 1 (and perhaps the most simple(?))
Hope someone can point me in the right direction.
/Jimmy