RESEARCH IN ADAPTIVE PATTERN RECOGNITION.
Final rept. 1 Apr 65-31 Mar 70,
INFOTON INC BURLINGTON MASS
Pagination or Media Count:
The report contains a review of the significant aspects of studies which have not been extensively reported elsewhere. The emphasis has been on the development and evaluation of pattern recognition schemes whose performance either does not depend or has minimal dependence on a priori knowledge of the underlying statistical behavior of the pattern classes to be discriminated. Significant developments include a nonparametric procedure which first transforms the unknown probability structure of a pattern discrimination problem to the real line and then partitions this induced space by a method based on the difference of empirical distribution functions a least-squares functional expansion about the Bayes optimum decision surface which leads to a type of polynomial discriminant function a transformation such that a class of stochastic processes can be characterized by the distribution of a set of discrete parameters and two procedures for ranking features according to their effectiveness in discriminating among dichotomous pattern classes. Author