ITU-T Rec. H.263 and Superimposed Prediction
Kluwer Academic Publishers eBooks · 2006
In our experiments, we observe that variable block size and superimposed prediction can be combined successfully. Superimposed prediction works efficiently for both 16× 16 and 8× 8 blocks. Multiframe motion compensation enhances the efficiency of superimposed prediction. The superposition gain and the multiframe gain do not only add up; superimposed prediction benefits from hypotheses that can be chosen from different reference frames. Superimposed motion-compensated prediction with two hypotheses and ten reference frames achieves coding gains up to 2.7 dB, or equivalently, bit-rate savings up to 30% for the sequence Mobile & Calendar when compared to the reference codec with one reference frame. Therefore, superimposed prediction with multiframe and variable block size motion compensation is very efficient and practical for video compression.