Measuring the Quality of Hypotheses in Model-Based Recognition

Model-based recognition methods generally search for geometrically consistent pairs of model and image features. The quality of an hypothesis is then measured using some function of the number of model features that are paired with image features. The most common approach is to simply count the number of pairs of consistent model and image features. However, this may yield a large number of feature pairs, due to a single model feature being consistent with several image features and vice versa. A better quality measure is provided by the size of a maximal bipartite matching, which eliminates the multiple counting of a given feature. Computing such a matching is computationally expensive, but under certain conditions it is well approximated by the number of distinct features consistent with a given hypothesis.