Software utilities

Google researchers announced at the International World Wide Web Conference held in Beijing the development of an algorithm to rank images, called Visualrank (pdf available). Google engineers claim this new image search results in much more relevant image results. This algorithm get the first images returned by a standard query and then analyzes them, finding the more similar ones, using a similar approach to the one used for Pagerank (the algorithm for ordering web page results). The paper shows image search results for some sets of images, such as Mona Lisa (on the left), grouped and linked (the two big ones on the center are the more relevant ones).

The old image ranking algorithm was based on text associated to images on the page, leading sometimes to inaccurate, or also completely wrong, results.

The algorithm has been developed by focusing on the top 2000 searched products on Google, such as iPod, XBox or Zune (the Microsoft MP3 reader).

This has not yet been developed, but you can have a look at the Google image search to see for example the face recognition feature, try it and check if it is enabled on your website.