December 19, 2006 1:00 AM PST
3D tech will search for just the right photo
- Related Stories
Heathrow testing biometric security checksDecember 6, 2006
Big Blue could monitor borders, shoplifters, mooseNovember 6, 2006
Search engine for personal files to be bundled in PCsSeptember 22, 2006
Spying an intelligent search engineAugust 18, 2006
Face recognition comes to photo albumsMarch 13, 2006
Start-up soups up surveillance camerasJanuary 25, 2006
IBM's 'Marvel' to scour Net for video, audioSeptember 29, 2004
The Swedish company essentially takes a two-dimensional photograph and extrapolates it into a 3D model. A computer then takes the 3D model and searches for photo matches on the Internet and in public photo sites like Flickr.
The company announced early Tuesday that it will start conducting field tests with select users in January and then launch a public beta later in the first quarter.
The 3D model, although virtual, allows a computer to filter out differences related to the lighting, the camera angle, and the angle of the subject's face in two or more photos. Such differences can otherwise throw off search results. By filtering out differences, the search results are more accurate, according to Jan Eric Solem, founder and CTO. Even if one photograph is a direct head shot, Polar Rose's search engine can turn up three-quarter and profile shots.
"We can create a 3D image from a single picture," Solem said. "Face recognition usually doesn't work because of lighting and pose. If you have 3D images, you can compensate for that."
Polar Rose is also working on a tool for searching for faces in video streams.
Image search has become a booming topic for companies and researchers, particularly as the number of photographs and videos posted to the Web grows. Many photo search services, such as Google's image search, actually don't search data within the images at all. Instead, the search engines look at the identifying data that accompanies a photo. A search on Jay Leno, thus, will turn up pictures of the comedian as well as any family pets named Jay Leno.
Other companies, such as EverNote, have created programs that can recognize words in images. Meanwhile, IBM and 3VR Security have come up with technology that searches on the images contained in video streams.
Earlier this year, a company similar to Polar Rose, called Riya, also launched a photo recognition service.
Although both accomplish the same task, the underlying methods for identifying and searching for matches differ between Riya and Polar Rose. Riya uses contextual recognition--the software "looks" at a person's face, but also considers the shirt, hair and other features in the photograph, on the principle that many of the same cues that appear in one photo of a person will crop up in a subsequent one.
Riya has also opened a subsidiary called Like.com that helps a person shop by searching for products with a similar appearance.
Polar Rose's algorithms do not focus on specific facial features. Instead, they examine the relative position of each prominent feature on one's face and compare that against the 3D model. (The company's name comes from a flower-shaped mathematical curve. It's not from Sweden's location on the globe.).
"As long as the variability is less between your different pictures than a picture of someone else, it will find you," he said.
Consumers will be able to download a plug-in to use the software. The plug-in will also act as a vehicle for delivering ads and links to commerce sites, which is how Polar Rose plans to make its money.
Last month, the company announced $5.1 million in first-round funding from Nordic Venture Partners.