Google’s VisualRank

The Register reports that two researchers from Google recently presented a paper called “PageRank for Product Image Search.” The new system uses image recognition software and also seeks to understand what is pictured in the image and assign it to a “visual theme”, similar to how the main Google index ranks websites and pages in themes. The images are also run through an iterative procedure based on Google’s PageRank computation so a numerical weight (or rank) is assigned to each image. This early version of the algorithm reportedly cuts irrelevant results by as much as 83%.

Google and other image search engines usually rely on the text around an image, it’s alt and title attributes and the file name for indexing and ranking purposes and although VisualRank is merely an idea at the moment it seems highly likely this system could be applied to the whole Google images database within the next few years.

Leave a Reply