A time-recursive interpolation filter for motion compensated prediction considering aliasing

In this paper a new interpolation filter for motion compensated prediction in a hybrid video codec is presented. It allows to interpolate an aliasing affected image signal more accurate than conventional interpolation filters like bilinear or Wiener filter. In contrast to conventional filters, the new time-recursive filter uses not only the image signal of the current frame, but also the image signal of previous frames in order to interpolate the current frame. The filter has been tested with synthetic and real sequences. The experimental results for the synthetic sequence show, that with conventional filters it is not possible to predict an aliasing affected image perfectly even if the estimated displacements perfectly fit to the sequence. Furthermore it is shown, that with the new filter it is possible to predict an aliasing affected image perfectly. For the used real sequence a coding gain of 0.8 dB is achieved compared to an MPEG-4 codec