| 摘要 |
During transform-based video compression processing, motion vectors, which are identified during motion estimation and then used during motion-compensated inter-frame differencing, are constrained to coincide with block boundaries in the reference data. Block-based motion vectors have components that correspond to integer multiples of block dimensions. For example, for (8x8) blocks, allowable motion vector components are (..., -16, -8, 0, +8, +16, ...). Constaining motion vectors in this way enables the resulting encoded video bitstream to be further processed in the transform domain without having to apply inverse and forward transforms. In particular, an existing input bitstream is partially decoded to recover the motion vectors and prediction error (i.e., dequantized transform coefficients). Because the motion vectors coincide with block boundaries in the corresponding reference data, motion-compensated inter-frame addition can then be performed in the transform domain to generate transform data for subsequent processing (which may ultimately involvere-encoding the transform data into another encoded video bitstream). Because motion compensation can be performed in the transform domain, the bitstream data can be further processed in the transform domain and without having to apply expensive and lossy inverse and forward transforms. |
