Figure 2.

Selection of best matches from multiple candidates, showing decomposition of a cluster with three reference (R) and two target (T) features, as well as connecting lines representing six candidate matches. Red matches (edges) have the lowest penalization score for each cluster at each iteration and are selected. Dashed lines are conflicting matches also containing the best-matched feature and thus are discarded. Blue lines are matches that initially are not the best but are not conflicting with the best match; thus, they can still be chosen in later iterations. In this case, two matches are formed from the original cluster after two iterations (R1–T1 and R3–T2).