How do search engines find and rank images?

[stweb]

images.google.com has been the number one ranked referring domain to one of my websites for the last 3 months.
On the other hand, I've tried Google and Yahoo for searching images from my site and I'm not impressed with the relevancy of the results. I get better rankings, from Google at least, when searching keywords.
So ... how do search engines find and rank images?
- looking for search words in the file name?
- looking at the alt text (many of my images don't have alt text)?
- looking at the html content and other standard search information on a page?
- or?
And ... which is the best way to help the search engine deliver relevant results?

ST.

[xan]

Hey there,

I am no expert in this at all, but the guys working on image retrieval from the vision lab here use Color, Texture and Shape Features as their variables. Then they will use artificial neural networks, decision tree learning, Bayesian learning and k-nearest neighbor classifiers for classification, and things like k-d trees for the indexing task. I know that things can be refined using user feed-back but I guess that's like everything else isn't it.

I guess it might now be dated but a few years ago I did a bit of work with Kahindu, an Image Processing Program which is java based. Maybe that can reveal something.

Google and AltaVista index image file format and metadata about the image file itself, but not the image content directly. These will also look at the words in the document to put that image in some kind of context.