VisualRank – new from Google
Image search based on PageRank apparently
An interesting article in the NY TImes looks at Google’s new take on image search. While image search is generally based on associated text the new so called VisualRank, is an algorithm for blending image-recognition software methods with techniques for weighting and ranking images that look most similar.
There is some mention of ‘authority’ sites as a basis for ranking in the article so it looks like the ‘Page Rank’ concept can be applied to images that are most representative of a particular query.
Even Google admits that the computing power to use this across all of their stored images would be ‘impractical’ so they focussed on a sub-set of popular searches and found they could improve relevancy by 83%. That sounds good but I am not sure what it means in the real world of image search.
Still, hats off for trying to bring this into the normal web arena, there is no doubt this is a tricky one so let’s see how they get on.