News 

The company aims to make its image search results as relevant as text searches, but is hampered by a computer's reliance on text cues to decipher the content of a picture.

"Although image search has become a popular feature in many search engines, including Yahoo, MSN, Google etc, the majority of image searches use little, if any, image information to rank the images," claims Google researcher Shumeet Baluja in a research paper.

"Instead, commonly only the text on the pages in which the image is embedded (text in the body of the page, anchor-text, image name, etc) is used."

Such dependency on text can throw up freak results, such as a magazine front cover of Monica Lewinsky dressed as Mona Lisa appearing when people

ADVERTISEMENT

search for the painting.

So the Google researchers have devised a new algorithm called VisualRank that looks for visual themes across a range of photos, with images ranked according to how similar they are to other images that contain that theme.

So, for example, a search for McDonalds may look for the famous golden arches, with those photos where the logo has been partially cropped or not made the main focus of the image ranking lower than those that do.

"The second challenge is that even after we find the common features in the images, we need a mechanism to utilise this information for the purposes of ranking," writes Baluja.

"Simply counting the number of common visual features will yield poor results. To address this task, we infer a graph between the images, where images are linked to each other based on their similarity. Once a graph is created, we demonstrate how iterative procedures similar to those used in PageRank can be employed to effectively create a ranking of images."

"We implicitly rely on the intelligence of crowds: the image similarity graph is generated based on the common features between images. Those images that capture the common themes from many of the other images are those that will have higher rank."

Baluja claims that when 2,000 Google employees were asked to rank the relevance of VisualRank results compared to standard image search, the new algorithm returned 83% fewer irrelevant images.