E. A. Samoilin

Article is devoted development of a method of nonparametric filtering of impulse interferences on digital maps in the conditions of complete a priori uncertainty rather brightness allocations of signals and the interferences, including solution of a collection of tasks of the image processing realized by model of a multifunctional neural network. The purpose of offered operation is improvement of quality of filtering of impulse interferences on digital maps in the conditions of complete signal-interference a priori uncertainty. The offered method of image processing is grounded on procedure of nonparametric (rank) nonlinear processing under the aperture, the form and which size, for the purpose of the best restoring of a signal in the spoilt pixels and the least signal alteration in the pixels which are not containing interferences, are defined by useful signal differences (boundary tags) in the given vicinity. Assignment of the sizes of the aperture is carried out on the basis of independent calculation of distances to boundary units in four directions concerning aperture centre (upwards, downwards, to the left, to the right). Thus gradient the algorithm of selection of boundary units is noise-resistant and differs from known for usage of the adaptive masks, allowing ignoring impulse interferences on their tentative estimations of space position on the map. The specified algorithms of estimation of position of impulse interferences, selection of boundary pixels, and also rank processing are realized by one multifunctional neural network with the memory for extract of values of necessary weight coefficients at transition from one task to another. The presented results of numerical researches show that the developed adaptive method of nonparametric filtering of impulse interferences allows raising quality of processing of digital maps in the conditions of complete signal-interference a priori uncertainty in comparison with known solutions. The offered adaptive method of nonparametric filtering of maps is easily enough realized in program and hardware sort, and can be used in many digital intelligence systems of the analysis of the visual data.