Graphics Reference
In-Depth Information
Tabl e 3. 1
Comparison of motion compensation block sizes supported in different standards
Video coding standard
Supported block sizes for motion-compensated prediction
H.262
j
MPEG-2 Video
16
16
H.263
16 16, 8 8
MPEG-4 Visual
16
16, 8
8
H.264
j
MPEG-4 AVC
16
16, 16
8, 8
16, 8
8, 8
4, 4
8, 4
4
HEVC
64
64, 64
48, 64
32, 64
16, 48
64, 32
64, 16
64,
32
32, 32
24, 32
16, 32
8, 24
32, 16
32, 8
32,
16
16, 16
12, 16
8, 16
4, 12
16, 8
16, 4
16,
8
8, 8
4, 4
8
3.2.4
Residual Quadtree Transform, Transform Blocks,
and Transform Units
As already mentioned above, for transform coding of the prediction residuals in
HEVC, a CB can be partitioned into multiple transform blocks (TBs). A transform
block represents a square block of samples of a color component on which the same
two-dimensional transform is applied for coding the residual signal. The partitioning
of a luma CB into luma TBs is carried out recursively based on a quadtree approach.
The corresponding structure, denoted as the residual quadtree (RQT), determines
for each luma CB at the root of the RQT a collection of luma TBs at the leaves
of the RQT in such a way that the union of corresponding disjoint luma TBs is
covering the whole associated luma CB. Figure
3.7
shows an example of a 64
64
luma CTB that is subdivided recursively into luma CBs and luma TBs along the
corresponding nested quadtree structures of coding tree and residual quadtrees. In
general, the partitioning of chroma CBs into chroma TBs is described by the same
residual quadtree. As will be described below, there is, however, one exception, for
which the splittings of the luma and chroma CBs of the same CU are not the same.
By allowing different transform block sizes, the residual quadtree transform
enables the adaptation of the transform basis functions to the varying space-
frequency characteristics of the residual signal. Larger transform block sizes,
which have larger spatial support, provide better frequency resolution. However,
smaller transform block sizes, which have smaller spatial support, provide better
spatial resolution. The trade-off between the two, spatial and frequency resolution,
can be freely chosen by the encoder control, for example, based on Lagrangian
optimization techniques. In the next subsection, the RQT structure is described in
detail followed by the description of its related parameter signaling scheme as well
as a brief discussion of a fast encoder implementation for RQT using an early-
pruning criterion.
