Weighed Mean Generalized Median of Strings

Abstract

Structural pattern recognition is characterized by the use of symbolic data structures such as strings, trees, and graphs for pattern representation. Consequently, the structural approach allows to model not only unary properties of patterns but also structural relationships between different patterns, or the parts of a pattern. However, compared to statistical pattern recognition, one of the shortcomings of the structural approach is the lack of a rich set of basic mathematical tools. As a matter of fact, the vast majority of all recognition procedures used in the structural domain are based on nearest neighbor classification using some distance function. By contrast, a large set of methods have become available in statistical pattern recognition, including various types of neural networks, decision theoretic methods, machine learning procedures, and clustering algorithms [13]. In this paper we put particular emphasis on novel work in the structural pattern recognition that aims at bridging the gap between statistical and structural pattern recognition in the sense that it may yield a basis for adapting various techniques from the statistical to the structural domain.