Google has unveiled a technology which it says will make the task of searching for images on the web far simpler and more effective.

The new tool will attempt to do for the voluminous number of photos and other images on the web what Google's search engine did for web pages: namely, rank them in order of importance.

Image search has presented a significant problem for internet companies, because search engines cannot yet decipher objects within images that would be instantly recognisable to humans.

The vast majority of what is called image search is in fact text-based, with engines scouring the text around an image - usually captions - to discern the content and in turn rate its importance to a given query.

Google's new engine, details of which were recently presented at a web conference in Beijing, will be one of the first to examine the content of images to determine their relevance to a query.

Unlike standard approaches to the problem, which attempt to figure out whether an object in question is, for example, a cow by comparing it to a standard outline of a cow, Google's tool will work by recognising a particular pattern or 'visual theme' in a series of images.

The relative importance of one image in the series is then assessed according to how similar it is to other images which are also found to contain the visual theme.

In a search for 'McDonalds', for instance, the engine would recognise the golden arches as a theme across a set of images, but whereas in some the logo would be crooked, rotated or "not the main component" of the image - for instance in a photograph of a McDonalds outlet - in others it would be the the main focus, for example in a cut-out of the logo.

The engine would then draw up a set of links between the images based on their similarity to others in the set, and rate more highly those which showed "the typical yellow M" that most people would expect when search for an 'image of McDonalds', a Google researcher said in a paper.

"By treating images as web documents and their similarities as probabilistic visual hyperlinks, we estimate the likelihood of images [being] visited by a user traversing through these visual-hyperlinks," said Shumeet Balujaz, a Google researcher, in a paper he co-wrote with a colleague from the Georgia Institute of Technology.

In another example, the researchers said that a famous cover of the New Yorker which showed Monica Lewinksy as the Mona Lisa fared less well in a search for 'Mona Lisa' because it was not as similar to images containing the 'theme' of the famous pose as were some others.

'PageRank for Image Product Search', as the new tool is called, is potentially one of the most significant search-based algorithms to be produced since 1997, when Sergey Brin and Larry Page, Google's co-founders, set out a way to rank a website's importance based on how many others linked to it. They called their algorithm PageRank.

In a trial which compared the results for the 2,000 most popular queries on Google Images, the new algorithm was found to reduce the number of "irrelevant images" by 83 per cent in comparison with those returned by the existing Google Images search, the researchers said.

The trial compared what the two engines deemed to be the most relevant images for a given query with answers given by a panel of 150 people.

The new engine will join a number of products attempting to simplify the process of searching through visual content, including Blinkx, which has rolled out an engine for searching video.

Two years ago Riya, a Silicon Valley firm launched Like.com, a service which can be used by shopping sites to match descriptions given by customers - for instance 'red heels' - with products from a catalogue.

Explore Tech & Web

• Personal Tech

• The Web

• Gadgets & Gaming