The latest innovation from Google,
VisualRank, was unveiled at the
International World Wide Web Conference last week. VisualRank is an algorithm which combines image-recognition software with techniques for ranking similar images, and is essentially PageRank for images.
Currently, images and videos appearing on search engine results are typically generated based on tags or surrounding text, such as an image title. Despite decades of effort, image analysis has remained a largely unsolved problem in computer science. “We wanted to incorporate all of the stuff that is happening in computer vision and put it in a Web framework,” said Shumeet Baluja, a senior staff researcher at Google.
Google’s research paper “PageRank for Product Image Search” is focused on a subset of the images that the giant search engine has catalogued because of the tremendous computing costs required to analyse and compare digital images. While has not disclosed how many images it has catalogued, it asserts that its Google Image Search is the “most comprehensive image search on the Web.”
To develop VisualRank, Google researchers focused on the 2000 most popular product queries on Google, for each of which they determined the top 10 images, gleaned in part from Google Image Search results. Google employees then created a scoring system for image "relevance", with net result of VisualRank search returning 83 percent fewer irrelevant images.
This announcement follows in the tracks of a San Francisco
ad:tech session on search engine optimisation two weeks ago, where the importance of incorporating alternative content, including images and video, into Google search results was emphasised by Google’s Aaron D’Souza.