A Girl with a Cat

The final question of the interview with Sangam Pant, vice president of engineering at Lycos (December Bits), looks naive in its ignorance of one of the great mysteries of nature: vision. Today, thousands of bright but frustrated researchers are working on computer vision and image understanding. Frustrated because what we do works only in simple toy problems, and we, like anyone else, don' t have a clue as to why we see, let alone how to translate this extraordinary gift into a dumb algorithm. And frustrated also because you asked with indifference if Lycos is "actually doing the equivalent of optica l character recognition on the image" by looking at the bit map and determining that "it's a girl with a cat." OCR can now be reasonably accomplished on a single chip. We humans have more than 20 billion highly interconnected neurons dedicated to v ision. You can figure out the difference for yourself.

Maurizio Pilu, Ph.D.
Digital Media Department
Hewlett-Packard Laboratories
Bristol, U.K.

I did not mean in any way to slight the research community. I chose the phrase "equivalent of OCR" because it was an analogy that many nonspecialists would understand. Mr. Pant seemed comfortable with the question, and as his response indicates, is well aware that such an undertaking is currently unfeasible. I asked the question because I do appreciate the magnitude of the task and wanted to make it clear that Lycos is not doing image analysis but rather is analyzing the links that point to the images, sounds, and video clips. -- Dave Andrews, news editor