High speed edge detection by sampling a time series with an orthogonal neural network

We introduce a high speed edge detection method where a time series is sampled using an orthogonal neural network, ONN, that is operating in an autoassociative testing mode. The training is done in near real-time. The testing is very fast since there are no calculations involving Gaussian functions, Laplacian operators or convolution. The speed of edge detection is further improved by combining a very simple rule-based expert system with our ONN. A Fourier analysis of an autoassociative ONN is presented. It is also shown that the ONN can improve its performance while on the job using a monitor/teacher system.

[1]  M. E. Ulug.  A single layer fast learning fuzzy controller/filter , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).

[2]  R. Wooldridge,et al.  Advanced Engineering Mathematics , 1967, The Mathematical Gazette.

[3]  D. Pollen,et al.  Phase relationships between adjacent simple cells in the visual cortex. , 1981, Science.

[4]  Christopher M. Bishop,et al.  Neural networks for pattern recognition , 1995 .

[5]  S. Lu,et al.  Texture segmentation by clustering of Gabor feature vectors , 1991, IJCNN-91-Seattle International Joint Conference on Neural Networks.

[6]  M. E. Ulug.  ANN bandpass filters for electro-optical implementation , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.

[7]  M. E. Ulug.  A rapid learning orthonormal neural network for signal processing , 1992, [Proceedings 1992] IJCNN International Joint Conference on Neural Networks.

[8]  Bernard Widrow,et al.  Adaptive switching circuits , 1988 .

[9]  M. E. Ulig.  Event representation in fuzzy neural networks with qualitative inputs , 1995, Proceedings of ICNN'95 - International Conference on Neural Networks.