Discrete distance and similarity measures for pattern candidate selection

Abstract Within the area of pattern recognition research, there are many examples where pattern families are decomposed into or composed from a set of basic elements. Regardless of their decomposition/composition order, each pattern within such a family can be considered as a repeatable combination of pattern primitives, and a discrete distance space and a corresponding similarity measure to facilitate the pattern recognition process can be defined. The definitions and properties relevant to the above discrete distance and similarity measures are presented. Both principles and an implementation method for pattern candidate selection within the discrete distance space are discussed.