Polar Rose is a Swedish startup that has been developing facial recognition technology for image search since it launched at the end of 2006.
It has recently updated the look of its image search facility and made the image recognition plugin available for Internet Explorer users, so we’ve taken a look to see how effective it is…
How does it work?
Unlike traditional image search, which relies on the metadata accompanying the image to return a search result for a particular person or place, Polar Rose takes a more sophisticated approach to the problem.
The company’s technology turns a 2D image, as displayed on a computer screen, into a 3D image which, after compensating for light and pose and other variations, can be used to search for related photos.
The service does still rely on tagging, and Polar Rose is using its beta testers to tag the millions of images out there.
With the browser plugin installed, users are prompted to tag pictures when they appear on the websites they are visiting.
In this case, on the BBC Sport site, the Polar Rose icon indicates that I can label Max Mosley or David Beckham:
Click on the icon and you can put a name to the photo, as well as search for other photos of the same person:
Searching Having tagged the picture of Mosley, if I search for the FIA president on the site, I am shown the picture I tagged, as well as related images:
The search interface has been tweaked and is nice and easy to use, displaying different groups of photos of the same person.
For example, searching for ‘George W Bush’ brings up 41 ‘stacks’ of photos of the US president. In some cases this is because the photos are of his father, or chimps, but it could be that the technology doesn’t recognise that all these photos are of the same person.
There is clearly a lot of tagging work to be done as well, as many searches return very few results or none at all, though most seem to be accurate.
The Polar Rose technology is impressive so far. As a pure image search engine it may be more accurate than Google Image Search, but it doesn’t index nearly as many pictures.
The site has been in private beta for over a year now and has only just expanded the number of beta testers on the site, so this should increase the number of photos indexed by the company.
Whether the site will be able to better Google Image Search is unclear, but there are other uses planned for the company’s technology.
It will be offered through APIs to allow partners, including photo sharing and news sites, to add the functionality, while the technology will also be used for contextual advertising.
Polar Rose launches photo search tool in beta
Like.com image search engine lauched by Riya
Swedish startup raises $5.1m for image search technology