Graphics Reference
In-Depth Information
1
2
8
3
4
8
16
7
5
6
9 10
11 12
13
1
2
7
13
14
15
16
14
15
3
4
5
6
9
10 11
12
Fig. 3.4
Example for the partitioning of a 64
64 coding tree unit (CTU) into coding units (CUs)
of 8
32 luma samples. The partitioning can be described by a quadtree, also referred to
as coding tree, which is shown on the
right
. The numbers indicate the coding order of the CUs
8 to 32
yielding a complicated syntax. Additionally, if the transform size cannot change
within a coding tree block, the encoder cannot well adapt to the local statistics in a
video picture.
In order to overcome these potential issues, another processing unit, called
coding unit (CU), has been introduced in HEVC. As illustrated for an example in
Fig.
3.4
, a CTU can be split into multiple coding units (CUs) of variable sizes. For
that purpose, each CTU contains a quadtree syntax, also referred to as coding tree,
which specifies its subdivision into CUs. Similarly as a CTU, a CU consists of a
square block of luma samples, the two corresponding blocks of chroma samples
(for non-monochrome video formats), and the syntax associated with these sample
blocks. The luma and chroma sample arrays that are contained in a CU are referred
to as coding blocks (CB). The subdivision of the chroma CTBs of a CTU is always
aligned with that of the luma CTB. Thus, in the 4:2:0 chroma sampling format, each
2
N
2
N
luma CB is associated with two 2
N 1
2
N 1
chroma CBs.
At the CTU level, a flag named
split_cu_flag
is included into the bitstream,
which indicates whether the complete CTU forms a CU or whether it is split into
four equally-sized blocks corresponding to square luma sample blocks. If the CTU
is split, for each of the resulting blocks, another
split_cu_flag
is transmitted
specifying whether the block represents a CU or whether it is further split into four
equally-sized blocks. This hierarchical subdivision is continued until none of the
resulting blocks is further subdivided. The minimum size of CUs is signaled in the
sequence parameter set, it can range from 8
8 luma samples to the size of the CTU,
inclusive. When the minimum CU size is reached in the hierarchical subdivision
process, no splitting flags are transmitted for the corresponding blocks; instead it is
inferred that these blocks are not further split. Hence, if a low-complexity encoder
is used that does never use coding blocks smaller than a particular size, the CU size
in the sequence parameter can be set accordingly, which avoids the transmission
of unnecessary splitting flags. In typical encoder settings, the maximum range of
supported CU sizes is exploited so that CUs ranging from 8
8 to 64
64 luma
samples can be used.
