point
Menu
Magazines
Browse by year:
Direct Decode
Wednesday, May 1, 2002
Kris Popat, a researcher at Palo Alto Research Center Inc. (know to most as the legendary Xerox PARC until it was spun off as a wholly owned Xerox subsidiary in January), tends to speak deliberately, in a rich, baritone pitch that makes him sound immediately both serious and authoritative. Indeed, interviewing him over the phone, one wonders if he could have been a voiceover artist or broadcaster. Luckily for PARC, where he has been leader in the field of digital image processing since 1997, Popat never considered a broadcasting career — although he did actually dream of a career in professional basketball. “I was the tallest kid in the class,” he says.

Popat’s college years at M.I.T pointed him in another direction: the study of digital signal processing, or the representation of audio or video information as a visual pulse. For the uninitiated, today’s audio systems employ digital signal processing, as do satellite and high-definition TV.

“I was getting more and more interested in information theory, data compression, and the fundamentals of representation of information, as well as transmission and storage,” Popat says. He wanted to continue working with images as well, he recalls. Some day, Popat’s change of heart may lead to universally easier access to digital interactive libraries, among other developments.

One of Popat’s areas of research is document image decoding, or DID. In the simplest terms, DID is an approach to text recognition that involves pulling content from document images, using algorithms designed to find what’s most useful about them.
At PARC, Popat’s DID research is largely focused on methods of recognition, compression, and retrieval that are particularly sensitive to machine-printed text, page layout formats, and the logical structure of the text. PARC emphasizes using “formal probabilistic models” that approximate the statistics of the language and domain of discourse, as well as typefaces and page layouts, and degradations in image quality resulting from printing and imaging. The models’ algorithms can then “decode” a page image so that the resulting interpretation is precise and, hopefully, useful enough to make potential applications for a variety of scan-based products a reality.

In keeping with PARC’s privacy policies, Popat is tight-lipped about what applications he is currently working on, but a recent paper he produced suggests DID could be used in the redisplay of text images on hand-held devices, like PDAs, as well in low-cost digital cameras. About the latter, the paper states: “One might use a pocket digital camera to take snapshots of selected pages while browsing books in a library, rather than having to carry the books over to a scanner or copy machine.” Certainly, this application’s more practical uses would include analyzing large collections of document images, just as with digital libraries (Popat acknowledges that libraries are an interest, as is the ability to use the technology in any way that advances learning).

Whatever Popat develops going forward, he says that he has no interest in forming his own venture to speed any products to market or otherwise commercialize them outside of PARC. Popat spent two years as a product marketing engineer at Hewlett-Packard in between getting his graduate and doctorate degrees, and he says he found the business world ill-fitting. Even though he got his Ph.D. just as the high-tech economy was heating up and startups were sprouting like wild dandelions, Popat was never much tempted to go the way of startups or become CTO of an Internet company. “As I was finishing school, I was approached by a number of startups and by more traditional companies. It wasn’t hard to turn down the work.”

Explains Popat, “We’re all motivated differently. The prospect of wealth isn’t unappealing, but it hasn’t ever affected large decisions in my life either.” In fact, says Popat, what he realized at M.I.T, having been exposed to a number of different, albeit similar, disciplines, was that his love of learning was the thread he wanted to continue weaving throughout his life. “It just occurred to me that maybe the thing I was already doing, research, was the right thing. Nothing else has challenged me enough intellectually.”

A little surprisingly, unlike many Ph.Ds to emerge from M.I.T., Popat never seriously weighed becoming a university professor. Though he has guest-lectured at Berkeley and Stanford, Popat says it’s “incredibly difficult to both do research and teach full or part time.” More importantly, says Popat, he decided early on that teaching would also limit the time he could spend with his wife and two young children. “You enjoy far more flexibility in the world of research.”

It’s good news for PARC that Popat is happily dedicating his life to finding new ways to extract useful information. Popat won’t comment on the age-old story about competing IT companies capitalizing on PARC technology because Xerox was too slow to move on its own team’s inventions. Of Xerox’s announcement that it had turned PARC into an independent company, he says only: “It continues to be a stimulating place to work.”

On that note, Popat announces that that he has some statistical models to attend to, and ends the call.

Twitter
Share on LinkedIn
facebook