Uncertainty in the registration between two images remains a problematic source of error in performing change detection between them. While a number of methods have been developed for reducing the impact of registration error in change detection, none of these methods are based upon a statistical characterization of the uncertainty in the estimate of the registration transformation. When utilizing a feature-point based registration algorithm, we can compute a Cramer-Rao lower bound (CRLB) on the estimate of the registration transformation based on an assumed covariance in the feature-point locations. This information can be used to predict the variance on the location at which pixels will appear in the registered image, which can be used to estimate the bias and variance introduced into the pixel intensities by registration uncertainty. Here, we use this information to improve change detection performance and verify this improvement with simulated and experimental results.

Date Published: 19 May 2015
PDF: 10 pages
Proc. SPIE 9460, Airborne Intelligence, Surveillance, Reconnaissance (ISR) Systems and Applications XII, 94600H (19 May 2015); doi: 10.1117/12.2176438

Published in SPIE Proceedings Vol. 9460:
Airborne Intelligence, Surveillance, Reconnaissance (ISR) Systems and Applications XII
Daniel J. Henry; Gregory J. Gosian; Davis A. Lange; Dale Linne von Berg; Thomas J. Walls; Darrell L. Young, Editor(s)