| 摘要 |
A decoding method comprises: specifying a search window, and search path for this window, in a first forward reference frame; for each motion vector MVO in the search path - where each MVO points from a current block to a reference block - determining a corresponding second motion vector MV1, pointing to a reference block in a second reference frame and a function of MVO; calculating a metric, e.g. using sum of absolute differences (SAD), for each pair MV0, MV1; and selecting the MV0 producing a closest value to the optimal value of the metric as a motion vector for the current block. Bi-prediction (mirror vectors), as shown, or only forward reference frames/vectors — Figure 4 — may be used. The motion estimation (ME) process can use sub-partitions of the input block (Figure 6). Traditional (coder side) ME or the decoder-side ME can be selected based on a rate distribution optimization (RDO) criterion (Figure 7). |
