Fuzzy non-iterative ARG labeling with multiple interpretations

Relaxation labeling is a category of methods that have been employed in the problem of scene labeling and correspondence analysis. Among these, deterministic relaxation methods are particularly useful due to their speed advantage. This paper compares the form and functions of existing deterministic relaxation algorithms, and attempts to show the problem with the existing methods and the cause of them. We propose a new fuzzy set theory based evidence combining formula for object centered attributed relational graph matching problem from which a new deterministic relaxation labeling method is derived. The algorithm succeeds in transforming the combinatorial optimization within large scale neighborhoods into a maximum weight matching problem, thus allowing efficient computation. A non-iterative (non-relaxation) labeling algorithm is also made possible by this formulation. We demonstrate the advantages of the proposed algorithms on a road matching problem and 2D scene labelling problems.