Fuzzy filter based on neural network and its application to image restoration

A novel fuzzy filter based on a neural network is proposed in order to remove salt and pepper noise on images. The filter proposed here is obtained as the sum of the input signal and non-linear transform on the output of fuzzy reasoning filter, and the non-linear transform is obtained by a trained neural network. Examples of processing actual images with salt and pepper noise are shown to verify the high performance of this filter. Moreover, the influence of the training signal on the filter performance is shown to be helpful for selecting the training signal.