TY - JOUR
AU - Gadetska, S. V.
AU - Gorokhovatsky, V. A.
PY - 2018/10/04
Y2 - 2022/05/21
TI - STRUCTURAL CLASSIFICATION IMAGES USING BAYESIAN DECISION MAKING
JF - Radio Electronics, Computer Science, Control
JA - RIC
VL - 0
IS - 2
SE - Progressive information technologies
DO - 10.15588/1607-3274-2018-2-10
UR - http://ric.zntu.edu.ua/article/view/142992
SP -
AB - Relevance. Ensuring the effectiveness and multifunctionality of modern computer vision systems requires the creation of a variety of<br />effective methods for intellectual processing of visual information. The development of systems of structural recognition is directly connected<br />with the construction of new effective methods as well as the need to create a mechanism for assessing the effectiveness of these methods for<br />specific applications of visual data. Bayesian decision theory is one of the tools, based on the statistical characteristics of structural data. The<br />calculation of a posteriori probabilities of assigning a description of a visual object to a set of etalons makes it possible to directly perform the<br />process of recognition as well as preliminary evaluate the effectiveness of procedures for comparing or calculating the relevance of descriptions<br />with respect to a specific application image database. Special attention is paid to the study of the structure of the set of descriptors, which<br />directly affects the functioning of recognition systems.<br />Goal. Investigation of the possibility and peculiarities of the application of the statistical recognition theory in the decision-making<br />mechanism and the evaluation of effectiveness in the form of the probabilities of classifying an object description as class. Comparison of the<br />results of computations with experimental computer modeling data.<br />Method. A method of recognition based on the application of cluster characteristics of the image base using the Bayesian decision theory<br />is proposed. The result of investigation is the creation of a mechanism for evaluating the effectiveness of procedures for calculating the<br />relevance of descriptions with respect to the application image database.<br />Results. The main result of the paper is the confirmation of the fundamental relationship between methods of comparison with etalons<br />and the statistical approach in pattern recognition with respect to structural descriptions in the form of a set of characteristic features of<br />images represented by a cluster description. The statistical approach based on Bayesian estimates, which is simpler in sense of estimated costs,<br />can be used for preliminary calculations of recognition efficiency without costly experiments on software modeling.<br />The effectiveness of the developed method for calculating probabilistic estimates for applied image bases is proved. The result of the<br />classification demonstrated the universality and correctness of the application of the method, each of the test objects in several of the<br />examined image bases was correctly recognized.<br />The obtained numerical results of the computations are compared with the experimental data of computer modeling.Conclusions. In the conducted research the method of structural classification of images on the basis of a cluster representation of the<br />description by means of Bayesian decision theory is proposed. The basic idea of applying the corresponding mathematical approach is in<br />assigning the analyzed object to an etalon that has the greatest value of a posteriori probability. The developed method provides a sufficient<br />level of discrimination of images, which was confirmed by the described calculations and simulation results, is offered. Mechanism for<br />evaluating the effectiveness of the analyzed methods of structural recognition within the framework of the applied image database has been<br />introduced.<br />The scientific novelty of the research consists in the synthesis of a new method of structural recognition of images and preliminary<br />estimation of efficiency by using the means of Bayesian decision theory and constructing classificatory solutions in the space of a clusteretalon.<br />The practical significance of the work is the obtaining of applied computational models for the application of the methods of structural<br />recognition and confirmation of their effectiveness in specific applied image bases.
ER -