1 July 1992 Hardware-implementable neural network for rotation-scaling invariant pattern classification
Author Affiliations +
J. of Electronic Imaging, 1(3), (1992). doi:10.1117/12.60033
Abstract
The design, hardware implementation, and simulation of a shift invariant pattern recognizer based on a modified higher order neural network (MHONN) is presented. When the MHONN is integrated with centroid calculation and logarithmic spiral mapping subsystems, translation, rotation around the optical axis, and scaling invariant pattern recognition can be achieved. The design objective is to deal with large-scale images with possible pattern deformation, noise, and highly textured backgrounds. Images are acquired with a 256 x 256 infrared sensor. We describe the theory of the MHONN, its hardware implementation, and simulation results.
Rafael M. Inigo, Catherine Q. Xu, Begona C. Arrue, Eugene S. McVey, "Hardware-implementable neural network for rotation-scaling invariant pattern classification," Journal of Electronic Imaging 1(3), (1 July 1992). http://dx.doi.org/10.1117/12.60033
JOURNAL ARTICLE
20 PAGES


SHARE
KEYWORDS
Signal to noise ratio

Pattern recognition

Neural networks

Binary data

System integration

Image processing

Intercontinental ballistic missiles

RELATED CONTENT

Making the most of 15k lambda 2 silicon area for...
Proceedings of SPIE (September 07 1998)
Optimal filters from prior filters
Proceedings of SPIE (April 04 1997)
Simultaneous segmentation and statistical label fusion
Proceedings of SPIE (February 14 2012)

Back to Top