Accession Number : AD0664218

Title :   A NONPARAMETRIC APPROACH TO PATTERN RECOGNITION. PART I. THE LOCALLY DISJOINT CASE.

Descriptive Note : Technical rept.,

Corporate Author : INFORMATION RESEARCH ASSOCIATES INC LEXINGTON MASS

Personal Author(s) : Owen,Joel ; Brick,Donald B. ; Henrichon,Ernest

Report Date : 25 NOV 1967

Pagination or Media Count : 22

Abstract : A mathematically rigorous procedure is developed which transforms the underlying unknown probability structure of a pattern discrimination problem to the real line. This transformed probability space is then partitioned using the fact that the locations of the relative extrema of the difference of empirical distribution functions will converge to the boundaries of the likelihood decision rule. In Part I, a method is proposed based on the locations of the relative extrema for discriminating between two disjoint pattern classes. It is shown that this procedure will produce perfect discrimination with probability 1. (When the classes are locally disjoint (defined in the text), perfect discrimination is possible with only a finite learning phase). In Part II this procedure is modified to include the non-disjoint case. (Author)

Descriptors :   (*PATTERN RECOGNITION, CLASSIFICATION), MULTIVARIATE ANALYSIS, INFORMATION THEORY, MAPPING(TRANSFORMATIONS), PROBABILITY, SET THEORY, THEOREMS

Subject Categories : Cybernetics
      Bionics

Distribution Statement : APPROVED FOR PUBLIC RELEASE