[Image-SIG] Categorisation of images

[David Otton]

Looking for some advice, hope somebody here can help me :

I want to group images by similarity, ideally into a tree structure,
so that similar images are near each other in the tree, and I can find
candidate images with few comparisons.

(I want to find the image that's most-similar to a candidate from a
large database (approx 35,000 at the moment). RMS works, but it means
I have to compare every library image against the candidate every time
I do a search.)

Can anyone suggest an approach, some URLs, an algorithm I could
research? I'm just looking for some high-level suggestions here, or
even an explanation of why it can't be done.

TIA,

djo