VEGGIE VISION

PETER PERROTTA

HAWTHORNE, N.Y. -- For seven years Dr. Jonathan Connell has toiled laboriously in IBM's Thomas J. Watson Research Center here, trying to invent a way supermarket cashiers could better identify fruits and vegetables.

He thinks he's got it.

It's called Veggie Vision.

After several modifications, Connell is now predicting that Veggie Vision is about one year away from debuting at a supermarket near you.

How's it work?

Quite simply, by taking pictures of the produce after it is placed on a scale at the front end, Veggie Vision then analyzes it and produces a color signature by breaking down the produce's hue, color saturation and intensity.

Veggie Vision works from a database that includes a color analysis of the 150 or so different fruits and vegetables the typical supermarket sells.

Thus, within an accuracy rate of 95%, Connell said Veggie Vision could tell the difference between cabbage and cauliflower, lemons and lingonberries, and Macintosh and Empire apples.

Since produce remains some of the only remaining items in the supermarket that can't be bar-coded, Veggie Vision, if successful, will take the guesswork out of the mix for cashiers.

"I think it's a neat system," Connell said. "It actually works. It's accurate and it's fast."

Connell said that during the course of the project's seven-year existence, several big name retailers and POS system manufacturers have shown great interest in Veggie Vision. Many, he said, have been invited in to IBM's research lab to take a look-see.

In fact, according to Connell, Montreal-based Optimal Robotics, one of the leading manufacturers of self-scanning POS systems, is very interested in Veggie Vision as well.

Henry Karp, president of Optimal Robotics, agrees.

"We think it's a good idea," Karp said. "We have been looking at it for two years.

"We went and saw it at shows. We played around with it. We like it. We think it's cool technology," Karp said.

Karp said he thinks Veggie Vision has a real future as a component in a POS system, especially a self-scanning system.

However, Karp said he thinks there are still some kinks that need to be worked out in Veggie Vision.

Karp said that for the most part Veggie Vision has operated in a "canned" environment. He thinks it needs real-world exposure.

"It's not deployed in the real world anywhere yet," Connell said. "We have talked to several companies who are interested. We've had internal field tests at IBM and other companies."

The next phase, Connell said, is to deploy Veggie Vision in the real world.

"We need feedback on what needs to be changed on it, if anything. Is it fast enough? Is it accurate enough?"

Another factor that will be important in the birth of Veggie Vision, Connell said, is if it can be integrated into a complete front-end system.

"Right now, it's a component technology. It's not integrated into anything."

The next generation of Veggie Vision will feature it integrated into a full front-end system, Connell said.

"I don't see any reason why it shouldn't be out on the market within a year," Connell said. "It's a very similar setup to a bar-code scanner."

Connell said that retailers and others who have seen the system have been impressed.

However, like any new thing, he said there are skeptics.

"The minute it [Veggie Vision] makes a mistake, they get upset," he explained.

In the end, according to Connell, it will all come down to value.

"It will be a question of where the retailers feel the value is," Connell explained.

One thing that is currently playing on the side of Veggie Vision is value. The early versions of Veggie Vision were expensive. Moreover, Connell said that most front-end systems, several years ago, were not equipped with enough computing power to handle it.

However, a lot has changed.

For one thing, Connell said, front-end technology has moved into the 21st century. Most major food retailers now have POS systems equipped with computer processors that can handle Veggie Vision.

Moreover, Veggie Vision itself has changed.

The early versions toyed with various set-ups that had more expensive cameras mounted either below or above the produce.

The latest version of Veggie Vision works with an inexpensive Web cam mounted above the scale where the produce is weighed, Connell said.

"Now, it [Veggie Vision] takes one photo from the Web cam," Connell said. "The produce is separated from the background, the plastic bags are removed [photographically] and then it is analyzed in terms of color."

Once the produce item is verified by Veggie Vision, an image of the item pops up on the cashier's screen for verification purposes. The cashier then verifies that it is that item and the produce is weighed and priced.

Connell said that in the early stages of development, Veggie Vision had a team of four researchers working on the project. However, over the past two years, Connell was the only one on it.

In its current state, Veggie Vision is complete, Connell said. He no longer will be working on it.

"We don't have any particular direction to take it right now," Connell said. "It is sitting there waiting."

Connell pointed out other intangibles that might make Veggie Vision an attractive option for supermarket retailers in the future.

He said Veggie Vision provides food retailers with a more accurate system of inventory control. He said unknowing cashiers who may be identifying produce incorrectly at the front end are giving retailers an inaccurate record of what is being sold.

He said with Veggie Vision, inventory control is presumably more accurate.

Moreover, Connell explained that by looking at the color of the produce, the computer can judge just how ripe the bananas really are.

"If it works well, it can be an integral tool for food retailers," Karp said.

"There is a big potential for it. It would especially be a good addition to self-scanning systems," Karp added.