Tsuhan Chen's work puts him at the nexus of several disciplines with one thing in common—visual data. His projects involve everything from capturing and enhancing the signal sent from a digital camera, to analyzing it for pattern recognition, to using it to render digital images of what the camera did not see.

It also involves how people perceive what they are seeing. "You really need to know the human end of this, too," he says. "It's fun research. I enjoy these kinds of things." 

Chen's interest in visual computing started with face recognition software for security applications—long before Sept. 11, 2001—and grew broader from there. "Now I'm more excited about consumer applications," he says. "We still work on face recognition, not for catching bad guys at the airport, but to help organize digital pictures." 

The software developed at Chen's lab uses several factors, including the distance between faces, to determine identities. The importance of the relative position of subjects became clear after statistical analysis of thousands of photographs. "Where people are helps determine who they are," says Chen. "It's almost like magic. Somehow, if you have a father holding a child as opposed to a mother holding a child, the spatial location is different."  

Chen, hired away from Carnegie Mellon to direct the School of Electrical and Computer Engineering, plans to balance his administrative duties with his research activities. The ten graduate students he's bringing with him are already working at full speed. "I do have very good students," he says, "and they are all very self-motivated." 

Another project they are working on could provide surveillance with fewer cameras and faster reaction times. The goal is to program groups of networked "robocams" to patrol buildings and historical sites. Simple image analysis tells them where there's movement, but it takes more sophisticated algorithms to identify suspicious activity. "We need to know what our target is," says Chen. "We want them to know when something is interesting enough to record, and second, we want them to work together to get the best image." 

If a couple of the robocams break down, for example, the others need to adjust to cover the gaps. Each robocam has its own computer, so deciding who goes where is a joint decision between parallel programs. "The group decision is made in a very distributed manner," says Chen. "Just like humans."  

Prof. Chen's Web site