Multi-window recursive adaptive neural filters

Abstract

Generalized adaptive neural filters are a class of nonlinear adaptive filters that includes stack filters as a subset. We further extend this class by using a multi-window approach. In this way we obtain a parallel recursive filtering operation and make better use of the implicit parallelism of the neural network architecture. The proposed neural network structure uses shared weight architecture for efficient implementation. Experimental results in actual image processing illustrate the efficiency of the approach.