Share on Facebook Tweet on Twitter Share on LinkedIn Share by email
Microsoft “Sees” People
By Jia Wu, China Internet Weekly
September 5, 2009 12:00 PM PT

Can computers eventually exceed the capacity of the human mind?

IBM's stock price rose by 3.6 percentage points the day after its computer "Deep Blue" defeated world chess champion Garry Kasparov in 1997. This triumph not only gave IBM an extra 200 million U.S. dollars, but it was also ranked as a top 10 news item by the Western media.

While The Matrix, The Terminator and other movies about man-machine conflicts seem to reflect concerns about computers of tomorrow, humans are constantly challenging their own limits through computational science. "In the past, psychologists and cognitive scientists believed that it would take a very long time for computers to match or exceed human minds, and many articles in respected publications attempted to prove that,” said Yi Ma. “Facial recognition was one of their most convincing pieces of evidence. It’s true that computers of that time did not do very well, but disbelieving articles have mostly disappeared since we took up this research."

Yi Ma was researching visual computation some nine months ago at a laboratory in the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. Still a tenured associate professor at the university, he now serves in a concurrent capacity as Research Manager of the Visual Computing Group at Microsoft Research Asia in Beijing. With a new platform at Microsoft Research Asia on which to work, Ma plans to challenge the limit of human vision — face recognition — through computational technologies.

Tight Schedules

Ma has a special identity at Microsoft Research Asia. He is not only the leader and public face of a project, but also an associate professor — a busy person in every sense.

He usually begins his day by replying to emails or calls from U.S. and giving a few instructions to his graduate students. He arranges routine affairs for members of the Microsoft Research Asia Visual Computing Group and deals with issues of project cooperation. At home in the evening, he continues with assignments from the U.S. or, on a few occasions, sits up late into the night for conference calls. "Working for Microsoft during the day, and calling the university at the night, I am actually doing two jobs for one salary," Ma joked.

Ma has not allowed a hectic routine to slow down the pace of his research. In his own words, teamwork minimizes his time spent on administrative issues, and on the other hand, "the most effective approach in management is to be an example to others. I have to do well in technological research, otherwise I will not be in a proper position to require younger researchers to take up missions that have great potential impact but are more risky.”

“Excitement” was the word that Ma chose to describe the difference between his current job and his previous one as solely a researcher. "I used to visit Microsoft Research Asia, and I felt it was the most active place of its kind. It was not until I came here that I realized how fast things are going around, and where its vitality originates," said Ma with the same excitement he was describing. "If I had continued to stay on campus, chances are that my vision would be limited. Working at a large company has brought me in contact with many different projects that have actually broadened my vision. Now I spend more time considering the applicability and future impact of my projects."

The parallel processing and multi-core processing capacity of computers today, along with innovative mathematical tools, have made computer vision technology applicable to practical products with actual impact. Ma has been thinking of ways to transfer his latest findings to applications.

“People Recognition”

The “most advanced finding in visual computing” Ma referred to, is "face recognition". By imitating human eyes and the brain, visual computing identifies objects, restores 3-D views and generates awareness and judgment. Face recognition identifies human faces by analyzing various facial expressions or partial images.

Ma used a demonstration to drive his point home. With the face recognition system activated, the video camera quickly captured Ma's funny facial expression and linked to the backstage face database. In a few seconds, the computer “locked on” to an image of Ma’s sullen face from a database of several hundred facial images. Few can imagine the series of complex algorithms behind this seemingly simple process. "Slight changes in the shape of eyes or the mouth are likely to make them unrecognizable, but our methods are far different from previous ones and have become registered patents."

Security might be the first field one thinks of when it comes to the application of face recognition technology. With an ordinary access control system, door cards can be transferred or replicated, which creates wide opportunities for a system to be breached. With face recognition, or other biological information identification features, security risks can be largely reduced. "Fingerprints, the iris, or DNA testing all disturb the user in some way; the advantage of face recognition is that precise monitoring is made possible without users knowing of its existence," said Ma.

Face recognition systems can be more broadly applied to personal information protection and on-line information retrieval. Even if a password is stolen, a face recognition system is still able to quickly tell whether the right person is using the computer. In the case of an imposter, the system can perform previously set procedures to shut down, and automatically photograph or identify the faces of illegal users. While at a meeting, if you forget the identity of the important figure you are talking to, you can take a picture of him or her with your cell phone camera, and the face recognition system will search the Internet for the person's name and background.

Technological developments have gone beyond graphic recognition and are heading for "people recognition" that enables terminals to have a type of human intelligence. A simple scenario: if you use face recognition technology to sort your family albums, the system could automatically help you classify and organize your pictures according to the shapes or expressions of the faces. This feature is already available on Facebook.

Exceeding Humans

"Technology is not only meant to exceed human capability, but also to challenge the human imagination.” We will be able to see each detail of a piece of artwork on-line, just as if we were viewing it at a museum — the lighting could even change in accordance with viewing angles. With a completely destroyed photo — even one that is covered in snow-like noise — a computer can clearly identify the photo’s subject, as long as a small part of the pixels maintain consistency. A tantalizing thought, yes?

Visual Computing is so amazing that it makes a world of human fantasy seem possible and suggests computing power that dwarfs that of all humans. By going beyond the limits of human beings, computers are enriching people’s daily lives.

"The more the Internet develops, the more its operations and content will be based on vision. Meanwhile, more and more of what is created by Internet users also revolves around vision," said Haiou Zhuang, NVIDIA Senior Marketing Director for the Asia Pacific Region. Bing, Microsoft’s new search engine, fully taps into the potential of visual computing for more precise and detailed image search, emerging as a new major threat to Google. Microsoft’s latest interactive games for the Xbox accurately restore real-time, 3-D images with cameras so that human-computer interaction resembles the interaction between people.

"The competitiveness of enterprises eventually depends on their talent reserve, or their ability to attract the best minds, am I right?" Ma smiled, adding that this clearly explains why Microsoft Research Asia was so determined to lure him from the University of Illinois. Laptop computers or mobile phones with face recognition systems will only reach their potential if they can get into the hands of everyday users. And that is what Microsoft Research Asia is hoping to do.