A NON-ITERATIVE PROBABILISTIC METHOD FOR CONTEXTUAL CORRESPONDENCE MATCHING

摘要

In this paper, we develop a framework for non-iterative structural matching using contextual information. It is based on Bayesian reasoning and involves the explicit modelling of the binary relations between the objects. The difference between this and previously developed theories of the kind lies in the assumption that the binary relations used are derivable from the unary measurements that refer to individual objects. This leads to a non-iterative formula for probabilistic reasoning which is amenable to real-time implementation and produces good results. The theory is demonstrated using two applications, one on stereo matching of linear features and the other on automatic map registration. The breaking points of the theory are also identified experimentally and the situations under which the proposed algorithm is applicable are discussed.