Cascade Non-linear Filters for Image Recovery
Elena Borisovna Solovyeva · Procedia Computer Science · 2019
Non-linear filtering is considered as a problem of cancelling non-Gaussian noise from the additive mixture of this noise and the initial signal. This task is formulated within the framework of the “black box” principle, when the non-linear filter operator is approximated by a behavioural model that establishes a relationship between the sets of input and output signals of a device. The case, when image signals distorted by the impulse noise excite a filter, is in highlight. Non-linear filtering results in the recovery of initial images with a assigned error estimated in the mean-square norm. A non-linear filter is synthesized as the cascade connection of the median filter, the cellular neural network and the Volterra filter. Under different impulse noise densities, the cascade non-linear filter is shown to ensure the higher accuracy of the image restoration in comparison with its singled out parts.