AN ACCOUNT OF OBJECT IDENTIFICATION CONFUSIONS

In 2 experiments, trained military observers identified vehicles in in frared (thermal) imagery that varied in distance, signal-to-noise rati o, and orientation. A measure of shape similarity was derived from a c ontingency tree that allowed prediction of the confusion rates between any 2 vehicles on the basis of the number of detectable, distinguishi ng parts. The mean confusion rates between pairs of vehicles were stro ngly correlated with the nodal distance between these vehicles in the similarity trees, even though the similarity trees had been constructe d without knowledge of the confusion rates. Such trees offer the possi bility for substantial improvements in the modeling of human object id entification and, when incorporated into training programs, offer a hi gh potential for reducing the likelihood of identification errors.