Search Menu
Photonics Media Photonics Buyers' Guide Photonics Spectra BioPhotonics EuroPhotonics Vision Spectra Photonics Showcase Photonics ProdSpec Photonics Handbook

Imager Uses Neural Network Algorithms for Machine Vision

Facebook Twitter LinkedIn Email Comments
Brent D. Johnson

Intelligent machines have long been the dream of systems integrators. The ability to install a component and walk away as it slowly sputters to life and begins learning its task has been the subject of science fiction for decades. The fantasy, however, has become reality.

Neural networks, which simulate the analog processing of human brain cells, are being applied in the machine vision environment, where pattern recognition of constantly changing elements demands complicated algorithms.

The zero instruction set chip invented by Guy Paillet of General Vision and IBM has been incorporated into Pulnix's ZiCam, enabling image patterns to be reduced to smaller components that can be independently modeled by separate neurons within a network. In other words, one neuron may receive information about the shape of the object, while another records the color. When the aggregate data is processed, it produces a memory of the object that includes each of these independent elements while retaining the entire image pattern.

Existing image recognition systems use normalized correlation or derived techniques that allow the fuzzy comparison between an inspected object and a single model. The ZiCam enables this comparison with 312 models in a few milliseconds. Furthermore, the real-time learning capability of the zero instruction set chip allows dynamic model addition and adaptation.

This is significant for machine vision because objects may be viewed under differing light conditions and in various orientations. The ability to recognize that an object is a chair, despite alterations in its appearance, will reduce the time that engineers must spend creating algorithms to compensate for these factors.

"The biggest issue here is that it is programming-free," Pulnix President Toshi Hori said. It can process new objects without new instructions because it learns as it goes. "You can treat it just like a human being. And the more you teach it, the more clever it gets."

Pulnix has several of the cameras working at beta sites and is almost ready for commercial release.

Photonics Spectra
Feb 2003
machine vision
Interpretation of an image of an object or scene through the use of optical noncontact sensing mechanisms for the purpose of obtaining information and/or controlling machines or processes.
Accent on Applicationsanalog processingApplicationsbrain cellsmachine visionmachine vision environmentneural networks

back to top
Facebook Twitter Instagram LinkedIn YouTube RSS
©2020 Photonics Media, 100 West St., Pittsfield, MA, 01201 USA, [email protected]

Photonics Media, Laurin Publishing
x Subscribe to Photonics Spectra magazine - FREE!
We use cookies to improve user experience and analyze our website traffic as stated in our Privacy Policy. By using this website, you agree to the use of cookies unless you have disabled them.