Global Image Alignment With Any Local Match Measure

The paper presents a new algorithm for parametric image alignment which when given any local match measure applies global estimation directly to the lo cal match measure data without rst going through an intermediate step of local ow estimation This algo rithm can be applied to any local match measure such as correlation normalized correlation squared or ab solute brightness di erences statistical measures such as mutual information etc Global estimation con strains the analysis of the local match measure thereby avoiding noisy local motion estimation Our algorithm bears resemblance to direct gradient based alignment methods but is more general as it is not restricted to minimization of brightness di erences only In particular when the generalized global alignment algorithm is applied with a normalized correlation match measure it results in a new global correlation based alignment algorithm which combines the robust ness and accuracy of global alignment with the broad applicability of the normalized correlation match mea sure We show via examples the satisfactory perfor mance of this algorithm even in di cult situations where the gradient based and ow based methods fail We also show how the algorithm can be applied to multi sensor image alignment