Identifiability: a fast way to measure identification performance

摘要

We present a new measure called target identifiability, as an efficient alternative for measuring identification scores. Identifiability is operationally defined as the amount of blur required to reduce the target signature to its identification threshold. It can quickly be determined using a simple adjustment procedure. To validate the new measure, we measured the identifiability of targets in a set of real and simulated thermal images. The identification scores for these targets were available from a previous study. Our results show that identifiability indeed determines identification performance. Sufficient accuracy can be obtained with only a few (typically 2 or 3) trained observers. The associated measurement procedure is simple and requires only a limited amount of time.