The Prospects of the Development of the Theory of Automatic Feature Generation in Pattern Recognition Problems

摘要

A new theory of feature formation in pattern recognition on the basis of stochastic geometry and functional analysis is presented. A new approach to the generation of features, independent of the motions or linear deformations of the images, is put forward. An important property of the considered set of features is the possibility of representing each of them as a convolution of three functionals.