Techniques are developed for determining spatial or geometric distortions between two images of the same scene. The first procedure is iterative linearized least squares estimation (LLSE) for determining small geometric distortions between images. Error variances for these estimators are derived which are interpreted as noise-to-signal ratios for translational and rotational registration. The natural measure of the signal strength of an image for translational registration obtained from these variances is used to establish threshold settings in a new algorithm for fast translational registration. This algorithm belongs to the class of sequential similarity detection algorithms (SSDA's) recently developed for translational registration. Finally, an implementation of an image registration system incorporating all these techniques is described.
Date of this Version