A recent article by The New York Times reports on a research paper presented at the International World Wide Web Conference in Beijing by 2 Google scientists, which described the development of 'VisualRank' by the search engine. This is the prototype algorithm that combines image-recognition software techniques with the methods for weighting and ranking images that look most similar. Described as the equivalent of Google's main PageRank system adapted for digital images, it's being seen as potentially a big step forward for the quality of image searching on the web.

Image search has long been one of the most popular 'vertical' search options on the main search engines and more recently Google, Ask and the others have been 'blending' image results into the main search results list, when relevant, thereby giving image search results far more exposure to mainstream searchers. Up until now, image results are usually generated from information in the file name and surrounding page content, but Google - and others - are working on ways to analyse image form and shape similarities.

The article says that Google has been working on a sample of around 2,000 of the most popular product queries from Google’s product search - so items such as iPod and Xbox - and then sorting the top 10 images both from its main ranking system and the standard Google image search results. Using a team of 150 Google employees, the researchers then created a scoring system for image “relevance”, which is being referred to as VisualRank. The outcome of this development, according to the researchers, was to retrieve new image results that were 83% less irrelevant.