Problems in The Design of Pattern Recognizers

Abstract

In this book a new method for the design of pattern recognizers, PRs, is presented. The heart of the method is a non-exhaustive, iterative procedure by which new effective pattern attributes can be generated from existing, less effective attributes. With this method each pattern is first reduced to a binary sequence by a suitable coding method. Each sequence is then described in a statistical sense by the observed frequency of occurrence of certain selected binary words. The categorization of the patterns is performed on the basis of such observed values. Any pattern attribute used with this method is the observed frequency of occurrence of some specified binary word. The method is therefore called: The Frequency of Occurrence of Binary Words Method, or the FOBW method. The method should also be of value to related disciplines where pattern recognition concepts are used, e. g., artificial intelligence and robotology.