Title :
Using a Hopfield network for rotation and scale independent pattern recognition
Author :
Hemminger, Thomas L. ; Raez, Carlos A.
Author_Institution :
Sch. of Eng. & Eng. Technol., Penn State Univ., Erie, PA, USA
fDate :
27 Jun-2 Jul 1994
Abstract :
This paper outlines a two-dimensional pattern recognition paradigm which is independent of rotation and scale. It is based on the Hopfield neural network, using a configuration similar to that employed for the traveling salesman problem. Enhancements to the original Hopfield design are included based on the eigenvalues of the connection matrix and through extensive simulations. The goal is to determine the underlying linear transformation between a binary valued test image and an unknown input pattern by minimizing the energy within the network. Experiment has demonstrated that this unconventional scheme performs successfully on a variety of rotated and scaled images and is robust against additive noise
Keywords :
Hopfield neural nets; eigenvalues and eigenfunctions; matrix algebra; pattern recognition; 2D pattern recognition; Hopfield neural network; binary valued test image; connection matrix; eigenvalues; linear transformation; rotated images; scaled images; Additive noise; Educational institutions; Eigenvalues and eigenfunctions; Hopfield neural networks; Intelligent networks; Noise robustness; Paper technology; Pattern recognition; Testing; Traveling salesman problems;
Conference_Titel :
Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Conference_Location :
Orlando, FL
Print_ISBN :
0-7803-1901-X
DOI :
10.1109/ICNN.1994.374711
