Global motion estimation using block matching with uncertainty analysis

The paper presents an approach to dominant global motion estimation which is based on estimation of displacements of a small set of blocks. For each block, a robust motion-compensated block matching measure is evaluated for a fixed range of displacements. Analysis of matching measure values provides displacement estimates, and the related uncertainty is also evaluated using a gradient based thresholding method. The results of the uncertainty analysis are then utilized in global motion estimation for outlier analysis and parametric motion model fitting. The performance of the technique is evaluated in experiments which show the usefulness of the approach.