mattg3 - 12:19 pm on Dec 20, 2006 (gmt 0)

Or maybe you could actually read the post first, since I said pages are still dropped despite having unique titles, unique descriptions and unique alt text. Good example of mimicing Google's behavior though!

The video quality isn't that good
but : Content-Based Image and Video Retrieval
[video.google.co.uk...]

They might be using a certain dataset [dunno if a talk equals likely implementation] that defines after a lot of iterations variables for a mathematical statistical model, posterior.

This is a statistical model that has an error margin. Sometimes this error is huge. Also if the model is wrong you will be off track.

I wouldn't wonder if they have several concurrent ways of trying to classify images and check out which one is the best.

Bayesian stats are "we have no clue what else to do stats". Sometimes they work excellent, sometimes they tell you nothing.