Motion compensated temporal filtering and motion vector coding using longer filters

Three-dimensional (3-D) subband/wavelet coding via a motion compensated temporal filter (MCTF) is emerging as a very effective structure for highly scalable video coding. Most previous work has used two-tap Haar filters for the temporal analysis/synthesis. To make better use of the temporal redundancies, we propose a MCTF scheme based on longer biorthogonal filters. We show a lifting based coder capable of subpixel accurate motion compensation.; If we retain the fixed size GOP structure of the Haar filter MCTF, we need to use symmetric extensions at both ends of the GOP. This gives rise to loss of coding efficiency at the GOP boundaries resulting in significant PSNR drops there. This performance can be considerably unproved by using a sliding window implementation in place of the GOP block. When we employ the 5/3 filter, its non-orthogonality causes PSNR variation, which we reduce by employing filter-based weighting coefficients.; Overall the longer filters have a higher coding gain than the Haar filters and show significant improvement in average PSNR at high bit rates. However the necessary doubling in the number of motion vectors to be transmitted, translates to a drop in PSNR at the lower bit rates. However since we have motion vector data available for every frame pair, we can use this temporal redundancy for more effective motion vector estimation. We then use the motion field at higher temporal resolutions as a starting point for the motion estimation at lower temporal resolution, and we use a smaller refinement range that reduces the complexity of motion search and also gives rise to a more uniform motion field.; For encoding the motion vectors a conventional approach was to follow a fixed quadtree scanning order and encode the motion vector prediction residuals along that path using the adaptive arithmetic is coder (AAC). We find that the motion vector coding performance can be improved by employing a context based binary arithmetic coder (CABAC). We can get more accurate motion vector predictions from (a) the neighboring blocks or (b) previous frames at the given temporal resolution or (c) from the motion field from the lower temporal resolution.; Finally, instead of unidirectional motion estimation, a bidirectional motion field is estimated for each temporal high subband and we reverse it to infer motion field for temporal low subbands. This helps to minimize the energy of the temporal high subbands and improves the coding efficiency.