The Brown & White

Computers are at once intelligent and painfully dumb. They can compute, catalogue, compress, delineate, display and direct in seconds.

But can they read?

Sort of. Researchers at Lehigh have spent years exploring the nature of the computer's ironic existence. Why is it that humans can create machines that are vastly more intelligent than their own kind, yet these machines are incapable of the basic perception and recognition skills of a newborn child?

Professors Henry Baird and Daniel Lopresti are taking steps to answer this question. Their research of OCR (optical character recognition) software is slowly bridging the gap between artificial intelligence and artificial perception.

According to Barid, OCR technology is centered around "trying to get computers to see things the way we do" - to be able to read, translate and understand written documents just like humans.

This is more difficult than it sounds, however, and OCR technology is far from perfect. As of now, OCR machines can read clean texts written in Western languages effectively. However, problems occur when documents are handwritten, dirty, noisy, old-fashioned or low-quality.

OCR machines work similarly to scanners; after a document is placed on the machine, the document then appears on the screen. However, unlike scanners, an OCR machine can then read the document, allowing the user to search the text as if it was a word document.

To see this technology in action, search a document on Google Books. That's OCR in action.

According to Barid, one of the main impetuses of OCR research is the potential to create a massive, searchable online database of the world's written texts - in essence, a super Google Books. This would enable people to electronically access a nearly unlimited supply of printed information.

The problem with doing this is a computer's perceptive skills are very limited. Barid points out that when people receive a hastily scribbled postcard they can, after several read-throughs, begin to make sense of what's written. Computers struggle with this, however.

The problem computers have with reading noisy, dirty or handwritten images is reflected by CAPTCHA, a method commonly used to deter hackers from creating fraudulent online accounts. Think of the last e-mail account you made. You probably had to type the letters you saw in a crunched, scrambled or jumbled image - this is CAPTCHA.

Baird said he likes researching OCR and CAPTCHA because it is like an arms race where he gets to play both offense and defense. Baird works on creating more advanced OCR software and more complicated CAPTCHAs.

OCR technology has uses outside of civilian life, as well. In fact, the majority of the funding for Baird's research comes from the federal government, which is extremely interested in OCR technology's potential to help both the military and defense communities.

"Think, if you just broke into some place where terrorists have been and there are all these documents. How do you make sense of that?" Baird asked. The answer seems to be through OCR.

Baird and Lopresti received large grants from the Multilingual Automatic Document Classification, Analysis and Translation (MADCAT) project, which focuses on creating software capable of scanning documents in other languages into English, so they can be quickly read and interpreted in the field.

The two professors also received money from the Defense Advanced Research Projects Agency (DARPA) and are currently working on the Document Analysis and Exploitation project (DAE), which aims to collect written data and research and enter that information into an online database where it can be searched, shared and expanded upon.

Despite the progress Baird and his fellow researchers made in developing OCR software, he still feels the technology is at least 50 years away from perfection. Although it is simple for humans to see part of something and assume the whole is there, computers struggle with this sort of abstract thinking.

Baird thinks this has to do with human's evolutionary necessity to assume the whole of something is there, despite only seeing a part of it. Recognizing a tiger in the bush despite only seeing its tail is the example Baird gave of this thinking, which seems simple to us, yet is difficult for computers.

However, OCR research nonetheless represents the first steps toward creating a computer that is both artificially intelligent and artificially perceptive. Researchers are on their way to creating a machine in our own image.

For now, however, Baird concedes computers are still both "extremely brilliant and extremely stupid."