Speeding up the generalized adaptive neural filters.

Abstract

A new class of adaptive filters called generalized adaptive neural filters (GANFs) has emerged. They share many things in common with stack filters and include all stack filters as a subset. The GANFs allow a very efficient hardware implementation once they are trained. However, the training process can be slow. This paper discusses structural modifications to allow for faster training. In addition, these modifications can lead to an increase in the filter's robustness, given a limited amount of training data. This paper does not attempt to justify use of a GANF; it only presents an alternative implementation of the filter. To verify the results, several simulations were performed by corrupting two images with varying amounts of mixture noise and Gaussian noise.