Motion compensation method and device using bidirectional optical flow
摘要
說明書
The sequences used in the experiment were 4 for Class A1 (4K), 5 for Class B (FHD), 4 for Class C (832×480), and 4 for Class D (416×240), and the experiment was conducted using all frames of the respective videos. The experimental environment was random access (RA) configuration, and the BD rates were compared by conducting the experiment by setting the QP to 22, 27, 32, and 37.
According to the present embodiment, BIO was skipped by about 19% on average, and 32% of BIO was skipped in Class A1 (4K), which has the largest amount of computation. The experiment showed that, as the resolution of the image increases, the ratio of skipping increases. The result of the experiment may be considered as significant because increase in resolution substantially increases the burden in terms of the amount of computation.
In addition, although there was an increase of Y BD rate of 0.02% on average, a BD rate difference of 0.1% or less is generally considered negligible. Accordingly, it may be seen that compression efficiency is almost the same even if the BIO is selectively skipped according to this example.
