Recognition algorithms based on the construction of threshold rules using two-dimensional representative pseudo-objects

¹ Research Institute for the Development of Digital Technologies and Artificial Intelligence, 17A, Boz-2, Tashkent, 100125, Uzbekistan
² Institute of Fundamental and Applied Research, TIIAME National Research University, 39, Qori Niyaziy str., Tashkent, 100000, Uzbekistan

^* Corresponding author: farhodtorg@gmail.com

Abstract

The development of recognition algorithms is discussed in the article; they are built using threshold rules based on representative pseudo-objects that provide a solution to the recognition problem in conditions of high dimensionality of feature space. A new approach is proposed, based on the formation of a set of two-dimensional base pseudo-objects and the determination of a relevant set of two-dimensional threshold proximity functions when constructing an extreme recognition algorithm. A parametric description of the proposed recognition algorithms is given, presented in the form of a sequence of computational procedures, the main of which are procedures for determining: 1) groups of tightly coupled features; 2) a set of representative features (RF); 3) groups of tightly coupled pseudo-objects in the RF subspace; 4) difference functions between objects in the two-dimensional subspace of RF; 5) groups of tightly coupled pseudo-objects in the RF subspace; 6) a set of basic pseudo-objects; 7) difference functions between the basic and simple pseudo-object in the two-dimensional RF subspace; 8) functions that differentiate between a pseudo-object and a class; 9) discriminant functions in the two-dimensional subspace of RF; 10) groups of tightly coupled separating functions; 11) basic separating functions in each group and 12) integral recognition operator for basic discriminant proximity functions. The results of a comparative analysis of the proposed and known recognition algorithms are presented. The main conclusion is that the implementation of the approach proposed in this study allows us to move from a given feature space to a space of RFs of lesser dimension.