Graphics Reference
In-Depth Information
Inter-16×16
Inter-16×8
Inter-8×16
Inter-8×8
Intra-16×16
Intra-8×8
Intra-4×4
Fig. 3.3
MPEG-4 AVC
for inter-picture coding (
top line
) and intra-picture coding (
bottom line
). If the Inter-8
Macroblock partitioning modes supported in the High profile of H.264
j
8 is chosen,
the 8
8 sub-macroblocks can be further partitioned into 8
4, 4
8,or4
4 blocks
8
8 luma samples. Each of the resulting sub-macroblocks can be coded as a single
8
8 block, using a single set of motion parameters, or they can be further split into
two 8
4,two4
8, or four 4
4 blocks. Note that the supported subdivisions for
8
8 sub-macroblocks are the same as the subdivisions for macroblocks. For coding
the luma prediction residual of inter-picture coded macroblocks, the High profile of
H.264
j
MPEG-4 AVC supports transform coding based on 4
4 and 8
8 blocks, the
chosen transform size is signaled on a macroblock level. If a macroblock is coded
in the Inter-8
8 mode and at least one of the sub-macroblocks is further subdivided,
the syntax element signaling the transform size is not transmitted but the usage of
the 4
4 transform is inferred; transform coding across boundaries of blocks used
for motion-compensated prediction is not supported. The chroma residual signals in
inter-picture coding modes are always coded using the 4
4 transform.
In HEVC, the basic processing units into which video pictures are partitioned
can be as large as 64
64 luma samples. A direct application of the H.264
j
MPEG-4 AVC macroblock syntax to the coding tree units in HEVC would cause
some problems. On the one hand, choosing between intra-picture and motion-
compensated prediction for large blocks is unfavorable in rate-distortion sense. In
P and B slices, typically most of the samples can be well predicted using motion-
compensated prediction. Only for a small amount of samples, intra-picture coding is
advantageous in rate-distortion sense. If the standard would allow to choose between
motion-compensated prediction and intra-picture coding only on the level of coding
tree units, it would result in a significant loss in coding efficiency. Actually, it has
been shown [
21
,
35
] that enabling the decision between intra-picture and inter-
picture coding at units smaller than a 16
16 macroblock can increase the coding
efficiency. On the other hand, for allowing such fine block structures for motion-
compensated prediction as in H.264
j
MPEG-4 AVC, the concept of macroblock and
sub-macroblock modes would have to be extended over additional hierarchy levels