Evaluation of Error Metrics for Image Registration of Multiple Images

Abstract

When two or more images of a scene are joined, one would like to know the accuracy of the joined image. In this paper we carry out a comparison of several different error metrics which are computed directly from the given images and their join (i.e., no "ground truth" value is assumed). We compare the error metrics by correlating them with a map registration error which we regard as a "ground truth" or "gold standard." The goal is to have a notion of reliability of the different error metrics which will be valid when ground truth reference is not available. This analysis should be a useful tool for quantifying aspects of image registration algorithms and for tuning the parameters of the registration algorithm.