Estimating Band-to-Band Misregistrations in Multivariate Imagery

In this paper, we use simulations and some theory to show that many of the standard techniques used to estimate band-to-band misregistrations in multivariate imagery are biased. These biases, although typically small, become important because they are much larger than the standard errors of the estimators obtained for most image pairs, which often consist of sample sizes of the order of 105 or 106. We develop an empirical method for effectively correcting the biases in some of the methods. Typically, this enables us to estimate misregistrations to within about 1/100th of a pixel.
This article is from 'Computing Science and Statistics: Proceedings of the Symposium on Interface Critical Applications of Scientific Computing (23rd): Biology, Engineering, Medicine, Speech Held in Seattle, Washington on 21-24 April 1991,' AD-A252 938, p587-592.