Sensor model appraisal for image registration

Registration, the alignment of reference and sensed images of the same scene, but taken at different times, viewpoints, or with different sensors is fundamental to numerous applications (e.g. image fusion, change detection, object recognition). A typical registration function is sensor model parameter estimation (e.g. sensor location and orientation) to optimize image match or alignment. We apply a global optimization method for this purpose, and adopt a Monte Carlo approach to quantify (appraise) sensor model parameter properties (e.g. covariances, marginal distributions). The latter are expressed as Bayesian integrals and are evaluated via importance sampling of models drawn from an approximation (derived from a Voronoi cell construct) to the posterior probability distribution. Our likelihood formulation is based on distance transform measurements. The latter are culled from the total available set of observations using selection criteria derived from robust variants of the standard Chamfer and Hausdorff distances.