Global motion estimation for image sequence coding applications

Global motion compensation improves the image prediction while adding only a small amount of side-information in an image sequence coder, thus saving bitrate and/or improving quality. A 3-parameter model is considered for global motion representation, including zoom ratio and horizontal and vertical pan. Two algorithms are based on block displacement estimates and take into account superimposed local motion. The first algorithm selects a set of most reliable block displacements and applies a least squares (LS) scheme to obtain an initial estimate of the model parameters. A Verification Stage is used to discard blocks whose displacements are not compatible with this estimate. Finally an LS scheme, using only the verified block displacements, is applied to obtain a finer estimate of the model parameters. The second algorithm uses a Hough transform to separate background blocks whose displacements are processed by a LS scheme. Simulation results are given for the `Table-Tennis´ image sequence containing zooming, and show a rate reduction of over 20%