ACCOUNTING OF THE SET OF RANDOM FACTORS WHEN USING THE MINIMEX CRITERION IN THE PROBLEMS OF OBJECTS RECOGNITION

In modern pattern recognition systems, the source data is usually random values, and the results of statistical processing of such data can lead to significant recognition errors. This article discusses the possibility of taking into account the errors of the source data in the case of using the minimax criterion. It is assumed that the descriptions of objects are a priori probabilities of the appearance of objects and the conditional probability density distribution of feature values, the parametric form of which is known. To determine the estimates of the parametric model, taking into account the errors of the values of functions and arguments, the methods of confluent analysis were used, allowing one to obtain unbiased estimates of the parameters. It is shown that taking into account the errors of the parameters of conditional probability densities of the probability distribution leads to the need to take into account the error of the boundaries of the separation of classes and the need to correct the formulas for determining errors of the first and second kinds. The error of separation of the attribute space leads to the emergence of a zone of uncertainty, the width of which will depend on the errors of the parameters of conditional probability distribution densities. In the article, a method is proposed for estimating the boundaries of separation of a feature space for normally distributed conditional probability distribution densities, taking into account the errors of the source data, in which the errors of the parameters are determined based on the variances of parameter estimates. The developed approach can be used in the tasks of object recognition by a variety of random signs.

1. Gorelik A.L., Skripkin V.A. Metody raspoznavaniya. – M.: Vysshaya shkola, 1984. – 208 s

2. Greshilov A.A. Matematicheskiye metody prinyatiya resheniy: Ucheb. posobiye dlya vuzov. – M.: Izd-vo MGTU im. N.E. Baumana, 2006. – 584 s.

3. Fedorov V.V. Teoriya optimal'nogo eksperimenta. – M.: Nauka, 1971. – 312s.

4. Fuller W.A. Measurement error models. New york ect.: – Wiley, 1987. – 440 p.

5. Kendall M., St'yuart A. Statisticheskiye vyvody i svyazi. – M.: Nauka, 1973. – 900 s.

6. Duda R., Khart P. Raspoznavaniye obrazov i analiz stsen. – M.: Mir, 1976. – 511s.

7. Gagarin YU.Ye. Sposob identifikatsii ob"yektov s uchetom sluchaynogo kharaktera iskhodnoy informatsii // Trudy mezhdunarodnoy konf. «Identifikatsiya sistem i zadachi upravleniya» SICPRO’2000. – M.: Institut upravleniya im. V.A. Trapeznikova RAN, 2000. – S.2087-2093.