Motion compensation using second-order geometric transformations

Motion compensation is a very effective process which can be used in video data compression for increasing the prediction accuracy in inter-frame coding. The conventional way of performing motion compensation as defined in the H.261 and MPEG standards, is to divide the “new” frame into a number of nonoverlapping rectangular blocks and then find the best match for each of these blocks by displacing a search window of the same size in the “previous” frame (or also the next frame for bidirectional prediction in MPEG). A block prediction based on this method is only accurate if the motion between the two frames is of a translational nature. There is no flexibility to change the size, rotation, or shape of objects if the motion that has taken place between the two frames contains such distortion. We evaluate the use of second-order geometric transformations (GTs) as a means of motion compensation. These transformations are capable of modeling the kind of changes mentioned above and therefore when used properly they have the potential of providing a much more accurate prediction during motion compensation