Adaptive weighted order statistic filters using back propagation algorithm

An adaptive weighted order statistic (WOS) filter is proposed. It can adaptively estimate the parameters of WOS filters according to its inputs and outputs. Since the number of variables of a WOS filter is equal to its window width, this adaptive algorithm is quite efficient. Another distinct advantage is that the adaptive WOS filter can proceed without use of threshold decomposition, which means that any discrete-time continuous value can be used as the input of the WOS filter. Some deterministic properties of WOS filters are discussed. A neural network structure is designed to realize this special stack filter. A learning algorithm is proposed to obtain the parameters of WOS filters. Some simulation results are presented to demonstrate the performance of the learning algorithm.<<ETX>>

[1]  Moncef Gabbouj,et al.  Optimal stack filtering and the estimation and structural approaches to image processing , 1989, Sixth Multidimensional Signal Processing Workshop,.

[2]  M. Atkins,et al.  Sorting by Hopfield net , 1989, International 1989 Joint Conference on Neural Networks.

[3]  Geoffrey E. Hinton,et al.  Learning internal representations by error propagation , 1986 .

[4]  Edward J. Coyle,et al.  Stack filters , 1986, IEEE Trans. Acoust. Speech Signal Process..

[5]  Richard P. Lippmann,et al.  An introduction to computing with neural nets , 1987 .

[6]  M. K. Prasad,et al.  Weighted median filters: generation and properties , 1989, IEEE International Symposium on Circuits and Systems,.

[7]  Edward J. Coyle,et al.  Stack filters and neural networks , 1989, IEEE International Symposium on Circuits and Systems,.

[8]  John W. Tukey,et al.  Nonlinear (nonsuperposable) methods for smoothing data , 1974 .

[9]  J J Hopfield,et al.  Neural networks and physical systems with emergent collective computational abilities. , 1982, Proceedings of the National Academy of Sciences of the United States of America.