Smart Cameras
Keep Their Brains on One Chip
By Toshi Hori

Just as human eyes capture images and the brain recognizes and processes the information, a camera and computer system together can provide artificial intelligence. This is, of course, the basis for machine vision systems, which captures images and sends them to a computer for analysis.

Modern machine vision uses complicated mathematical algorithms to calculate whether a part that is under analysis matches- within some range of tolerances- a programmed "ideal." The technology works, but researchers are working on ways to speed up and "smarten" machine vision systems so that they operate more like the human brain's biological neural network.

The concept of the neural network goes back to Mac Culloch and Pitts in 1943. They introduced the concept of biologically interconnected neurons and demonstrated the ability to compute arithmetic and logical functions.

Since then, numerous studies and developments have been aimed at implementing the concept electronically. Despite such attempts, most solutions have been software-heavy, requiring too much his-speed computing power to make them practical for machine vision applications.

The latest developments by IBM France and Silicon Recognition Inc. that put the radial basis function and K-nearest neighbor functions into a silicon chip have opened up the potential of significant progress toward applying the neural network principle in industrial applications. The chip is called zero instruction-set computer processor: No instruction or programming is required to implement the arithmetic and logical functions.

Continued on Next Page