This paper proposed an edge detection method using Hopfield neural networks and 2D evolutionary operators. The algorithm maps a detected image into a Hopfield neural network in such a way that each pixel corresponds to a neuron, and utilizes a population of Hopfield neural networks simultaneously. Different Hopfield neural networks have the same weights but begin to update with different initial neuron output states. In order to resolve the local minimum problem inherent in Hopfield neural networks and enhance the exploitation ability of evolutionary operation in extreme large search space, the dynamic equation of Hopfield neural network and 2D evolutionary operators are carried out alternatively during network's update procedure. The experiments have illustrated its good performance.

Date Published: 9 October 2000
PDF: 4 pages
Proc. SPIE 4221, Optical Measurement and Nondestructive Testing: Techniques and Applications, (9 October 2000); doi: 10.1117/12.402641

Published in SPIE Proceedings Vol. 4221:
Optical Measurement and Nondestructive Testing: Techniques and Applications
FeiJun Song; Frank Chen; Michael Y.Y. Hung; H.M. Shang, Editor(s)