Google researchers are cracking the nut that is image search using the company’s much-maligned PageRank computation that it uses to assign Web sites scores based on relevance.
Google programmers Yushi Jing and Shumeet Baluja, using so-called “authority” nodes on an inferred visual similarity graph, propose an algorithm to analyze the visual link structure that can be created among a group of images.
In an approach the programmers are unimaginatively calling VisualRank, a numerical weight is assigned to each image to measure its relative importance to the other images being considered. So how is this different from how other search engines parse video?
The researchers said commercial search engines tend to rank images based solely on the text clues of the pages in which images are embedded, ignoring the content of the images themselves.
“The problem of answering a query without image processing is that it can often yield results that are inconsistent in terms of quality,” Jing and Baluja wrote in a white paper delivered at the World Wide Web conference held in Beijing April 21-25.
For example, check out the difference in results between the queries for the Eiffel Tower and McDonald’s (PDF).
No word on when we can expect some mass implementation of VisualRank, but the company is understandably pumped about the breakthrough.
A Google spokesperson told me, “We are very excited about the VisualRank technology, which provides a better way of ranking image results based on the visual content within an image. It marks an important step forward in integrating many novel machine learning, statistical and computer vision approaches towards providing better image search results to our users.”
I agree with that assessment. Google, Microsoft and Yahoo are all working to provide some comprehensive search called universal search, but they can’t deliver on that vision unless they stop using the text in searches to determine results on images.
That just doesn’t make sense, and with more users consuming images and video, reliable image search is a competitive differentiator. For Google, VisualRank is the next step and could reinforce the company’s position as the premier search engine going forward.