Fuzzy Relaxation Labeling Reconsidered

The problem of assigning labels to objects given some relational constraints is occuring quite often in the field of pattern recognition. Among the different solutions proposed there is a particularly successful one: Relaxation labeling. Relaxation labeling has turned out to be a useful iterative scheme for parallel information integration forcing neighborhood constraint satisfaction. Consequently there has been done an abundance of work on relaxation labeling since the seminal paper of Rosenfeld et.al.[1]. There the authors discussed various models for relaxation labeling: discrete, probabilistic, non-linear probabilistic and fuzzy. Since then many new relaxation labeling schemes have been proposed only a few of which are incorporating methodologies from fuzzy set theory. It is shown that this is partly due to the traditional, but questionable classification of relaxation labeling schemes. A new classification is introduced which clarifies the role of fuzzy relaxation labeling. The ne..