Face-Off: New Software Recognizes Faces on Web
Dec. 20, 2006 — -- Let's say you've uploaded your holiday photos to a picture-sharing Web site, and you want to mark all the ones that include your cousin George.
Or you're doing some research, and you want to find pictures of a young Bill Gates.
Or, perhaps, you wonder about a long-lost love, and you'd like to see where he or she has gone since you broke up.
A Swedish startup company, Polar Rose AB, has now announced a face-recognition program that it hopes will make the countless millions of photos on the Web more searchable.
And it's free; the firm hopes to make money through advertising.
"Social networking on the Web is very popular, and very visual," said Mikkel Thagaard, vice president for business development at Polar Rose. "It's not usually searchable."
Finding Faces; Solving Mysteries
The solution offered by Polar Rose uses the kind of face-recognition software that, at a more detailed level, is used by law enforcement agencies and governments trying to track down terrorists.
Such software was used by Canadian police trying to solve a murder mystery that dates back nearly 40 years.
Using a face-recognition program, Ontario Provincial Police said on Tuesday that they'd identified a body that had been found in 1968 in a rural area northwest of Toronto.
Matching the body to old photos, they announced the victim was Richard Hovey, a 17-year-old musician who had disappeared from a Bohemian Toronto neighborhood in 1967.
Hovey's relatives released a statement saying they were "very relieved to be able to bring our brother home after years of anguish."
Modern-Day Uses: Making Web Pictures Searchable
Polar Rose says its software is similar, but the company will not sell it to law enforcement agencies.
"It's a tool for making photos searchable and sortable," Thagaard told ABC News. "We're very interested to see how people will want to use it."
Here's the idea: If you go to the image-search feature on Google, Yahoo, or other major search sites, you can enter words to describe what you're after, but you have to count on those who posted the pictures to label them properly.
A Google image search for "Charles Gibson," for instance, will turn up plenty of pictures of Charlie -- but also a 1909 sketch of a young woman in Edwardian dress. Why?
Because it's by Charles Gibson -- Charles Dana Gibson, the artist whose "Gibson Girl" drawings defined feminine beauty a century ago. Someone at the University of Houston, assembling a Web page about Charles Dana Gibson, dispensed with his middle name in a caption.
If that's what happens when you look up someone as well known as Charlie, what about your best friend from college? Or what if you're trying to cull through the pictures you've posted on MySpace or Flickr for friends to see? Most of the images on the Web don't have meaningful captions at all.
The software from Polar Rose would let you input a picture of the person you're interested in, and the software would then analyze it -- superimposing a grid over the face to determine the shape of the mouth, the depth of the eye sockets, the slope of the nose, and so forth. Then, when you used it to search the Web, it would identify faces with the same attributes.
Many, of course, will be close but no cigar. The software can be thrown off by shadows, or unusual expressions, or odd angles. You'll have to go through the results, labeling the ones that were right.
"The user has a great deal of impact on the results," Thagaard said.
But once you've given the picture a label, other users of Polar Rose's program would see it as well. Web users, as a community, would help catalogue pictures.
Will it work? Mike McGuire, an analyst at Gartner Inc., a research firm, isn't sure.
"You really have to have scale," he said. "It doesn't work if you have a few thousand people using it. It only works when you have a big, big crowd."
He went on, "The wisdom of the crowd is a very powerful thing. It's not a solution for everything."
The Associated Press contributed to this story.