Hybrid neuro-fuzzy filter for impulse noise removal

A new filter is presented for images which are highly corrupted by impulse noise. The proposed operator is based on a hybrid neuro-fuzzy approach. The network structure of the filter is specifically designed to detect differe nt patterns of noisy pixels typically occurring in highly corrupted data. T he fuzzy mechanism embedded in the network aims at removing noise pulses wi thout destroying fine details and textures. A learning method based on the genetic algorithms is adopted to adjust the network parameters from a set o f training data. Experimental results show that the neuro-fuzzy filter is a ble to yield a very effective noise cancellation and to perform significant ly better than state-of-the-art operators in the literature. (C) 1999 Patte rn Recognition Society. Published by Elsevier Science Ltd. All rights reser ved.