Spatio-temporal compression of the motion field in video coding

In conventional fixed size block motion compensation (FSBMC), the motion information is a set of motion vectors called motion field. There is a great amount of correlation within the FSBMC motion field: spatial correlation between vectors of the same frame and temporal correlation between vectors of consecutive frames. This correlation is the potential for a more efficient coding of the motion. Variable size block motion compensation (VSBMC) and region-wise motion compensation (RWMC) methods were proposed to reduce the spatial correlation of the motion field, by merging several blocks having the same motion into one larger block/region and coding only one motion vector. Observing the temporal correlation of the motion field, a different technique was proposed to predict the motion of the current frame based on the motion of the previous frames. In order to use all the potential for the efficient coding of the motion field (motion field compression), both temporal and spatial correlation are used in the proposed method. The experimental results show a significant decrease of the motion information bit-rate