Graphics Reference
In-Depth Information
Tabl e 3. 5
Coding efficiency improvement for successively increasing the maximum depth of
the residual quadtree (RQT). The shown average bit-rate savings are measured relative to a
configuration in which the RQT depth is equal to 0. All supported CU, PU, and TU sizes are
enabled
Residual quadtree depth
Entertainment applications
Interactive applications
RQT depth equal to 1
0.7 %
0.7 %
Maximum supported RQT depth
1.2 %
1.0 %
Tabl e 3. 6
Coding efficiency improvement for successively increasing the CTU size and the
number of hierarchy levels in the coding tree. The shown average bit-rate savings are measured
relative to a configuration with a CTU size of 16
16 and a minimum CU size of 8
8 luma
samples
CTU size and minimum CU size
Entertainment applications
Interactive applications
32
32 CTU, 16
16 minimum CU
9:2 %
17.4 %
32 32 CTU, 8 8 minimum CU
12:1 %
20.2 %
64
64 CTU, 16
16 minimum CU
12:7 %
23.8 %
64
64 CTU, 8
8 minimum CU
14:9 %
25.5 %
is subdivided for the purpose of signaling intra prediction modes. The results are
summarized in Table
3.5
. It can be seen that providing the possibility of applying the
transform across PU boundaries yields a bit-rate saving of 0.7 % for both application
scenarios. By allowing more than one level of the residual quadtree, the coding
efficiency can be further increased, even though the improvement is small compared
to the impact of other coding options.
In the next experiment, we compared different configurations for the CTU size
and the minimum CU size. As reference, we used a configuration with a CTU size
of 16
16 luma samples and a minimum CU size of 8
8 luma samples. This
configuration provides similar partitioning modes as the High profile of H.264
j
MPEG-4 AVC. The bit-rate savings obtained by successively increasing both the
CTU size and the depth of the coding tree are summarized in Table
3.6
. A significant
gain is already achieved if all block sizes are basically increased by a factor of 2 in
horizontal and vertical direction, corresponding to a CTU size of 32
32 luma
samples and a minimum CU size of 16
16 luma samples. By further increasing the
depth of the coding tree and the CTU size, the compression performance is further
improved for the tested HD sequences.
If we look at the development of video coding standards from H.262
j
MPEG-2
Video to HEVC, one key aspect for improving the compression performance was
to increase the set of supported block sizes for motion-compensated prediction
and transform coding. In the last experiment, we evaluated the associated coding
efficiency improvement using the HEVC reference software. It should be noted
that we do not compare the different video coding standards, but only investigate
the impact of increasing the supported set of block sizes for motion-compensated
prediction and transform coding. For all other aspects, the coding tools of HEVC
are used in the experiment. As a reference, we used a configuration that is
