Accession Number : AD0674317
Title : THE OPTIMUM METHOD OF PATTERN RECOGNITION FOR MUTUALLY CORRELATED CHARACTERISTICS (OB OPTIMALNOM METODE OPOZNANIYA OBRAZOV PRI VZAIMNOKORRELIROVANNYKH PRIZNAKAKH),
Corporate Author : FOREIGN TECHNOLOGY DIV WRIGHT-PATTERSON AFB OHIO
Personal Author(s) : Turbovich,I. T.
Report Date : 16 NOV 1967
Pagination or Media Count : 14
Abstract : In various papers dealing with the design of recognition machines the undimensional distribution functions are established after sampling, then assuming the characteristics are truly independent, the multidimensional distribution function is found. In this case the machine applies the minimization of error probability. In practice, however, it is in general impossible to establish a sufficiently complete system of uncorrelated characteristics. Consequently, the present author discards the statistical approach. The optimization criterion of the recognition machine is sought in the minimum of the number of standards needed for comparison (i.e., the simplicity of the recognizing device) rather than in the minimum of the error probability. The minimum of the number of standards is found by means of the so-called association function. Each point in the space of the characteristics is described by the totality of standard-associating functions. The standard-associating function of an arbitrary pattern is positive if the point of the space of characteristics under consideration is closer to that standard than an arbitrary realization of another pattern; otherwise, the associating function is equal to zero.
Descriptors : (*PATTERN RECOGNITION, *INFORMATION THEORY), ALGORITHMS, OPTIMIZATION, SAMPLING, CORRELATION TECHNIQUES, MATRICES(MATHEMATICS), USSR
Subject Categories : Cybernetics
Distribution Statement : APPROVED FOR PUBLIC RELEASE