Neural network application for primary local recognition and nonlinear adaptive filtering of images

An approach to neural network (NN) application to image local recognition based on several statistical parameter evaluations in a scanning window and their further joint analysis is put forward. Ways to deal with the images corrupted by dominant multiplicative or additive noise are discussed. The NN learning and structure selection methodologies are considered. The neural network classifier performance is analysed for the training and verification data sets. The possible applications of primary image recognition results, in particular, for further nonlinear locally adaptive filtering of images are proposed. The corresponding numerical simulation results proving the efficiency of considered techniques are proposed