Parametric Motion Vector Derivation for Video Coding

Video encoding methods save motion vector fields locally in a buffer to predict the following frames, thus reducing the size of the error pattern that is needed to decode the frames. This so-called motion-compensated prediction derives the motion vector field for the motion of pixel blocks from frame to frame. If several pixel blocks have identical motion vectors, it is sufficient to save only one of these pixel blocks. In this way it is possible to further reduce the amount of necessary data by 4 to 5 %.

The presented method reduces the amount of data even further, by using a model with 8 parameters of a higher order. We have to consider, that motion vectors which can be derived from a so-called warping matrix are not transmitted separately. Here the error rate decreases with the increase of the number of parameters. The decoder uses the motion parameters to reconstruct the vector field. Remaining motion vectors are transmitted without additional manipulation. Therefore, the amount of data that is transmitted can only decrease, an increase is not possible.