A novel nonlinear filter using layered neural networks

This paper proposes a novel realization of nonlinear filters suitable for the edgepreserving smoothing of an image degraded by a mixed noise environment composed of the Gaussian and impulsive noises. This filter consists of a layered neural network and a median filter. By using layered neural networks, the parameters of the proposed filter can adapt itself to the various noisy environments through the learning of a training image. The training method of the parameter of response functions is also proposed. These parameters have important effects for the performance of the proposed filters. An example is shown to illustrate the utility of the proposed filter.