On Mon, 16 Feb 2009 12:52:06 +0000, Tony  wrote in
<49996175$0$510$bed64819@news.gradwell.net>:

>No idea how it works, but the basic free search engine does work (I've 
>used it successfully in the past).  The major issue with the engine 
>isn't technical, it's quantity (it hasn't indexed enough images).

Quantity is the big issue if you want to search all the images on the
web. It means doing for images what those well known search engines do
for text and the scale and continuous updating issues are very similar.

>Their non-free tools suggest they can find parts of pictures, and 
>pictures which have been changed, and they show some examples.
>
>Personally, considering it's essentially a function of maths, I don't 
>doubt it's possible.  I've seen graphical effects I considered to be 
>near-magical in nature, so I'm not surprised by much technically image-wise.

The magic search terms are "near-duplicate image detection" and
"sub-image retrieval". This will turn up research papers such as
"Efficient Near-duplicate Detection and Sub-image Retrieval"
<http://www.intel-research.net/Publications/Pittsburgh/101220041248_261.pdf>

Given the dates of the papers I found it seems to be an active research
area.

-- 
Owen Rees
[one of] my preferred email address[es] and more stuff can be
found at <http://www.users.waitrose.com/~owenrees/index.html>