Google unveils new prototype image search system “VisualRank”
Google has revealed a new page ranking system for images called VisualRank. The biggest search engine in the world today uses PageRank and this determines a websites value, based on its content and is scored from 0-10. The higher the websites PageRank, the higher the site appears on Google’s listings for the related keyword if all things where equal, title, h1 number of page etc etc…
It is different with images, as rankings largely rely on analysing the text near the image, the images file name and the ALT text associated with the image. Whilst progress has been made in the field of face detection for images, other more traditional objects like stereos and food items remain largely untouched.
VisualRank works using image recognition software, which groups images together that seem most similar to each other. It then estimates “visual hyperlinks” which in turn produces a final ranking.
Google have said that they concentrated on 2000 of the most popular search queries to make the prototype, including iPods, Xbox and Ps3. It then sorted the top 10 images from both its PageRank system and from Google Image search results, after that they created a scoring system for an images relevance. Researchers have estimated an 83% drop in irrelevant images returned by the search system.