It Saved Them $100 Million the First Year They Used It

by Sargur N. Srihari

Getting a computer to read handwriting like a person would is an incredibly difficult problem. Many people can't even read their own handwriting at times.

As a graduate student I was interested in getting computers to understand patterns and images and things like that. There are so many ways you can go at it. From the visual angle, what is it that makes us recognize shapes? Some portions of the problem are highly mathematical. The question puts together the left and right sides of the brain, I suppose. It's a nice problem and satisfying to me. I approached the post office for funding and they funded the research.

After ten years, the post office said: "Look, we've funded this research of yours and we don't have anything out there to show for it. Can you now take a break from your research for a year or two and make a use for it?"

So I did. We ended up with a nice product. We developed an automated system that will read handwritten addresses off of postal envelopes. We're able to read about 30 percent of the addresses correctly, addresses which would otherwise be handled by human labor. It really works. The post office said that they never had a situation where they could take something out, put it out there and it really worked.

It saved them $100 million the first year they used it. So they said, "We believe your research is of value and we'll continue to support you to do more of this."

Over the years, hundreds of graduate and undergraduate students have been supported by the U.S. Postal Service and others in the funding of this and similar projects at our Center for Excellence in Document Analysis and Recognition (CEDAR).