The further development of cellular neural networks (CNN) based on multi-valued and universal binary neurons (MVN and UBN) has been carried out.
The following results have been obtained: 1) A new class of the effective non-linear filters, so called multi-valued filters (MVF) has been developed and templates for implementation of the new filters using CNN-MVN have been obtained. Comparison of the new class of filters with order-statistics ones shows that MVF are more effective for removing of the Gaussian, uniform, speckle and mixed (with impulse component) noise. MVF also may be successfully used for amplification of the high and medium frequencies (this operation leads to extraction of the image details) of MVF. They are more effective than linear filters used with the same aim. Templates for implementation of the corresponding filters using CNN-MVN have been designed, and their high efficiency for solving the problems of noise reduction, image enhancement and extraction of details has been proven; 2) A MVN-based neural network which is much simpler than a Hopfield one for solution of pattern recognition problems by neural analysis of the low part of orthogonal spectra has been developed. It has been used successfully in experiments with face recognition; 3) A solution of the precise edge detection problem, and edge detection by narrow directions has been carried out. Such an solution is reduced to the evaluation of the values of non-threshold Boolean functions. An implementation of the corresponding algorithms using CNN-UBN has been proposed; 4) A problem of orthogonal spectra extrapolation (or super-resolution problem) via approximation of the spectra, and its correction using multi-valued filtering has been solved.