Can someone explain how does the SI_PositionalColor descriptor works ?
I understand that the image is divided in rectangle by nxm
For each rectangle the most significant colour is represented in the descriptor.
Is that correct? What is the most significant colour?
Not an answer but this may help
A query like
WHERE image_id = 1
will produce nine SI_color values (m and n are both set to 3 by default), e.g.
I've done some experimentation on a simple image of a square grid of 3X3 distinct colours and I can't work out how positional colour is derived.
I expected, but didn't get, an RGB colour for each of the distinct colours assuming each colour has equal significance. I wonder whether there's some weighting taking place (e.g. more significance given to the centre of the image).
Please report any progress.