BEIJING--At Tsinghua University here, researchers have devised an application that allows security cameras to identify individuals. It's currently used to expedite the border-crossing process between Shenzhen and Hong Kong.
Hewlett-Packard is now trying to bring the application to consumer PCs for searching video files.
The Palo Alto, Calif.-based computing giant is opening a lab at Tsinghua, China's premier technological university. The facility, HP's first full-fledged lab built at a university, will initially focus on pattern matching and searching technologies for audio tracks, videos or photos. Ideally, the collaboration will result in applications for HP computers that will better wed its printer business to PCs or let HP offer capabilities that competitors won't have.
"There is definitely a need for this. Printing is about context. There is nothing to print if you can't find what you are looking for," said Patrick Scaglia, chief technology officer of imaging and printing systems at HP.
One set of applications from Tsinghua, for instance, can find songs in a music library by their rhythm characteristics, genre or vocal track. Type in "classical" on a genre search, and the software ferrets out the symphonic pieces on a hard drive through the sound in the track, not text tags associated with music files.
Another application lets users find songs by humming a few snatches of the tune.
Tsinghua also has a photo-matching application. Conduct a search on a face, and it will pop up matches in a hard drive by searching on the contours of the person in the photo. The application is currently used by some police agencies in China, according to Fang Chi, an associate researcher in the Department of Electronic Engineering at Tsinghua.
In an HP computer, the software, potentially, will let users click on a family photo, and the computer would pop up other ones. Similarly, users could conduct text searches on terms such as "birthday party" and get all the pictures on their hard drive in which large groups of people are surrounded by balloons. With these tools, conceivably, consumers wouldn't have to tag photos to allow them to be searched on photo-sharing sites.
Although the Tsinghua software is still somewhat experimental, it works pretty well, according to Scaglia. The current photo-matching application has been tested against a database of a million photos.
"We've searched the world, and it is the best one we've seen," he said. "It is quite compelling. It is a research thing. It is not ready to ship (on PCs), but it gives us confidence that the problem can be cracked."
In my own informal test of the video tool, it seemed to work, up to a point. A single image of me was inserted into the database. A roving security camera was able to pick me out and identify me in a group of three or four people. When I took my glasses off, however, the system labeled me as a "stranger," or unknown individual.
But in a more controlled environment like a border crossing, it seems to work pretty well, according to Chi. There are already 2 million faces logged in the databases at the Shenzhen-Hong Kong border, and officials there have been using it for a year. By using the application rather than the traditional methods of border control, the flow of individuals passing through the somewhat busy crossing is accelerated.
Tsinghua has been working on pattern recognition software for several years. The university has licensed software to cell phone makers and others that let users take a picture of street signs and then conduct Internet searches on the words captured in the photo, said Zhisheng Niu, vice dean of the School of Information Science and Technology at the university.
Under the arrangement, HP gets a license to use software and inventions developed in the lab. HP is also working with the university on management tools for ChinaGrid, a computing grid that connects 20 of the grids at major universities into a larger computing unit.
Collaboration between universities and private-sector companies, though around for years, is on the rise. Universities need money, and many companies no longer have in-depth labs for fundamental research.