The "similarity" of two images is a very complicated concept, when people are asked to tag images the overlap rarely goes above 20% on average (e.g. check out http://images.google.com/imagelabeler/ and try your hand at it). Think about it: you may think at the object level (both images have cars), concept level (both are happy images), color, etc. This is why large online stock image sellers still rely on tags extensively.

As a rough analogy, consider a textual example: Find a sentence similar to "It is a truth universally acknowledged, that a single man in possession of a good fortune, must be in want of a wife." Now, if you enter this sentence in Google, it retrieves documents that contain it, in TinyEye fashion. What other similarity is desired? Should it retrieve essays on Austen, on marriage, 17th century English literature...?

If you are interested in image similarity search, check out the Pascal challenge (http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2011/inde...). The advances in the last 5-6 years on object detection and visual feature extraction (which image similarity relies on) is amazing.

Here's an interesting article about image hashing and fingerprinting algorithms: http://www.hackerfactor.com/blog/index.php?/archives/432-Loo...

Everything from that blog is interesting, do you or anyone else got any suggestions for similar blogs?

Actually the presentation showed a query for an old picture returning images taken from the same place. It's being said that they're doing it by matching images' basic lines and shapes.