A simple generalized neuro-fuzzy operator for efficient removal of impulse noise from highly corrupted digital images

A generalized neuro-fuzzy (NF) operator for removing impulse noise from highly corrupted digital images is presented. The fundamental building block of the operator is a simple 3-input 1-output NF filter. The operator is constructed by combining a desired number of NF filters with a postprocessor. Each NF filter in the structure evaluates a different pixel neighborhood relation. Hence, the number of NF filters in the structure can be varied to obtain the desired filtering performance. Internal parameters of the NF filters are adaptively optimized by training by using a simple artificial training image that can easily be generated in a computer. Simulation results indicate that the proposed operator outperforms popular conventional as well as state-of-the-art impulse noise removal operators and offers superior performance in removing impulse noise from highly corrupted images while efficiently preserving image details and texture.