Discrimination between random signals and impulse noise in images which is optimal according to the criterion of minimum combined losses

Abstract

An algorithm is proposed for discriminating between random image signals and non-Gaussian impulse noise with a uniform brightness distribution which is based on a new optimality criterion combining the minimization of weighted unconditional probabilities of wrong decisions of the first and second kind and maximization of the probabilities of making correct decisions. Results of numerical studies of the proposed and the well-known Bayes algorithms are given showing that the use of the new algorithm reduces the discrimination error under conditions of complete a priori uncertainty of the image signal distribution.