People
Mapping the Technological Edges
Graduate student’s research gives computers the ‘vision’ to make advances in medical and security imaging
July People Notes
Mapping the Technological Edges
Graduate student’s research gives computers the ‘vision’ to make advances in medical and security imaging
By Julia Keller
Shahan Nercessian wants to give computers the eyes to see the world as humans do and the brains to interpret what they see.
An electrical and computer engineering graduate student, Nercessian works in the field of what is broadly called “computer vision”: enabling computers to aid people in making important security and medical decisions by recognizing and interpreting objects—for example, a suspicious package at an airport or invasive tumors in the body. Nercessian works to fine-tune mathematical algorithms so that computers recognize objects consistently across all kinds of images, from satellites maps to MRI scans.
“If you combine computer vision with robotics, it will be a great advance in medicine,” says Shahan Nercessian. Photo: Joanie Tobin
How we see is obviously a complex issue. “Humans don’t understand how we know what we know,” says Nercessian, E07, E10. “We know that this is a desk, and this is a stapler, and this is a computer,” he says, pointing to objects in his office. “Why is this still a mouse when it’s turned upside down? These kinds of issues pose huge problems for computers.”
The algorithms that Nercessian writes help a computer to determine the edges of an object, much like a person would trace an object’s outline. Nercessian calls up a black-and-white image of a woman wearing a feathered hat onto his screen. He then runs an edge-detection image-processing program. The computer instantly returns a complementary image that looks like paint-by-numbers: the edges of the woman’s hat and the features of her face are all outlined in white on a black background.
“You get the gist of what’s going on here in the picture, but the complexity of the information is drastically reduced,” says Nercessian. “The algorithm gives you structural information, and from there you can quickly figure out what’s going on with this object.”
Nercessian works at Tufts’ Simlab, the hub of signal processing and image-enhancement research directed by Karen Panetta, associate professor of electrical and computer engineering. Her research and work with the startup company BA Logix has produced image-processing technology that is leaps and bounds faster than many algorithms on the market today. For image processing in, say, an airport, which has lots of people carrying lots of things, the goal is to process images as quickly as possible with a high enough degree of accuracy to alert security agents to objects that are even remotely suspicious.
“Algorithm performance is oftentimes measured subjectively by the human eye,” says Panetta. “Algorithms like the ones that Shahan writes help us convert subjective perception into objective analysis.”
Nercessian now wants to fine-tune how algorithms process the image data, rather than focus on the speed of returning an image. “Usually, you’d like to take both speed and accuracy into consideration,” says Nercessian, “but you have to realize what the technology is for.”
Refining the algorithm to produce the sharpest images possible is key for biomedical applications. For a doctor or surgeon studying an MRI scan for malignant tumor growth, “image processing feeds into your decision making,” says Nercessian. “If you looked at it yourself, you wouldn’t have actually known that there was cancer. But if you see the processed image, you would say, ‘Obviously, it’s there.’ ”
Nercessian says his father, an electrical engineer who worked at Philips Electronics in the patient monitoring group, was a big influence in his decision to go into the biomedical side of engineering. “After taking my first image-processing course, I decided to focus more on image processing, particularly for medical applications,” says Nercessian. “I wanted to know, in a more direct way, that what I was doing was helping other people.”
The “crazy, out-there, futuristic goal,” says Nercessian, is sharpening the image processing enough so that computers could perform difficult surgeries without the aid of physicians. “Right now, doctors are under a lot of pressure to work as accurately as a machine does to perform [procedures] correctly,” he says. “If you combine computer vision with robotics, it will be a great advance in medicine.”
This story appears in the Summer 2009 issue of Alma Matters magazine.
Julia C. Keller, the communications manager at Tufts University’s School of Engineering, can be reached at j.keller@tufts.edu.
