Page last updated at 08:30 GMT, Wednesday, 14 May 2008 09:30 UK

Solving the web's image problem

By Angela Saini
BBC News

ESP screengrab
Google have bought a licence for the ESP Game

Every year, video game enthusiasts fritter away billions of hours in fruitless fun, earning little more than sore thumbs and a sense of satisfaction.

But to Luis von Ahn they represent an opportunity not to be missed.

The computer scientist at Carnegie Mellon University in Pittsburgh developed a game in 2003 that uses gamers to actively improve the web.

"We're trading entertainment for labour," says Prof von Ahn.

His Extra Sensory Perception (ESP) Game reveals the same image to two players and asks each to guess what the other person has written to describe it. If they agree, that word or phrase is then used to annotate the picture.

Repeating the same image with other pairs of players, the computer eventually builds up a detailed label.

100m pictures have already been labelled in this way, prompting the popular search engine, Google, to buy a licence from Professor von Ahn's team to create its own version of the game.

Google Image Labeler was launched in 2006 and helps to return better search results for online images.

Popularity problem

Tools like this are needed because as a general rule, computers do not understand images. Someone searching for a picture of a rabbit on the web, for example, will only find it if it has been correctly labelled with the tag 'rabbit'.

Unfortunately millions of online pictures and videos are either poorly tagged or have misspelt descriptions, making it almost impossible to track them down using search engines.

Luis von Ahn

If we outline enough pictures of dogs, for example, the hope is that in the long-run the computer will know itself the outline of a dog

Luis von Ahn

The ESP Game could be a quick solution to this problem. Professor von Ahn claims that if it were as popular as other online games, all the images on the web would be labelled in a matter of weeks.

The only barrier is making it interesting enough to attract enough players.

The game has 1m registered users, but only a fifth of them play it regularly. "There's no game out there that is liked by 100% of the population," he admits.

In an effort to get more players on board, his team is trialling four new games through a slicker, more user-friendly site called gwap.

One of them, called Squigl, asks a pair of players to trace around an object in a picture using their mouse, earning points for how closely they matches their partner's outline.

"This really trains the computer to see. If we outline enough pictures of dogs, for example, the hope is that in the long-run the computer will know itself the outline of a dog," he explains.

Another game, called Matchin, asks players to pick the most beautiful image out of two they are shown, helping to build a database of the web's most attractive pictures.

Semantic support

Games like this may improve web searches, but the ESP Game does not solve the underlying dilemma of getting computers to understand pictures for themselves, according to some researchers.

Jonathan Hare from the University of Southampton, for example, is working on the root of the problem.

He is using semantic web technologies to help computers identify what objects are in a picture and what they mean.

The semantic web is widely seen as the next step for the world wide web and involves information that is stored in machine-readable formats.

"One of the things we need for computers to recognise pictures is good data, and the ESP Game is a good way to get that data," he says.

But he is a not a fan of the game himself.

"I tried it and gave up after a few images. There is no incentive to keep playing unless you really enjoy it."

Print Sponsor

Google expands its search empire
18 Feb 04 |  Technology
Photo tool could fix bad images
08 Aug 07 |  Technology

The BBC is not responsible for the content of external internet sites

Has China's housing bubble burst?
How the world's oldest clove tree defied an empire
Why Royal Ballet principal Sergei Polunin quit


Americas Africa Europe Middle East South Asia Asia Pacific