Vol. 7, issue 06, article # 14

Abstract:

Within the framework of the Bayes approach to constructing decision rules for image recognition in the space of definitions of high dimensionality, we solve the problem of retrieving multidimensional conditional functions of the density of classes, based on normalizing transformations. These transformations satisfy two conditions. The first of them demands that estimated distributions of a separate feature agree with actual single-dimensional distributions at a practically acceptable reliability. The second condition states that the approximating distribution should describe statistical relations between the components of the vector of observations.