Simple Pattern Recognition

[SMurf]

Hello,

In another one of my bizarre hobby programs I'm trying to match sections of a 24-bit colour image to one of a number of 1-bit patterns.

Currently the steps I'm taking are:-

1. Reduce image colour depth to 8-bit greyscale
2. Increase pattern depth to 8-bit (essentially 0 and 255)
3. Compare the sums of the Euclidean distances between source and pattern pixels.

The thing is, I don't think that's right. I shouldn't be blindly checking colours against colours, because the colours are irrelevant. What I should be checking are that the edges of the colours roughly match up and maybe introduce a heuristic so that close matches with a few pixels out are still acceptable. Does that sound correct?

Also, three of the patterns involved are essentially all the same bit (0 or 1) so I should classify them and hold off testing against them unless everything else is a bad match, yeah?

[Salem]

Mmm, what about
- changes in luminosity of the captured image,
- changes in position (translation)
- changes in rotation
- changes in scale (zoom)

All of these will mess up a simple bit-for-bit attempt at comparing the image.

I'm gonna assume that you've googled already.