APPLICATION OF UNCERTAIN AND RANDOM VARIABLES TO KNOWLEDGE-BASED PATTERN RECOGNITION

摘要

The paper is concerned with recognition problems based on relational knowledge representations with two kinds of unknown parameters: uncertain parameters characterized by certainty distributions given by an expert and random parameters described by probability distributions. After a short presentation of uncertain variables and their application to the knowledge-based recognition under uncertainty, the different versions in two formulations of the recognition problem are discussed. In the first formulation two unknown parameters (uncertain and random) in the relation describing the set of object to be recognized are considered. The second formulation concerns the case with one unknown parameter in the knowledge representation, described by a certainty distribution with a random parameter or by a probability distribution with an uncertain parameter. Simple examples illustrate the presented approach. The application of so called C-uncertain variables and the description of the recognition system with three-level uncertainty are included.