Graphics Reference
In-Depth Information
CUs. Furthermore, it should be kept in mind that the transmitted depth values do not
necessarily correspond to the number of subdivisions when subdivisions are forced,
due to resulting TB sizes greater than the maximum transform size, as already
mentioned above.
Since the RQT structure is nested in the coding tree, it has to be signaled for each
coding tree leaf node, i.e., for each CU after transmission of CU prediction mode,
PU partitioning, and PU related syntax, provided that the residual signal of the CU is
represented by one or more non-zero transform coefficients levels. In that case, the
syntax element
split_transform_flag
is transmitted for every RQT node
to indicate whether it is a leaf node (with a value of 0) or an internal node (with
a value of 1). Note that for the cases presented in the previous subsection, where
a signaling of this flag would be redundant,
split_transform_flag
is not
explicitly signaled but inferred at the decoder side instead.
In addition to the actual structure of the RQT but often also as its replacement, it
is signaled whether there are significant, i.e., non-zero transform coefficient levels
present in a particular TB or the whole CU. For CUs using motion-compensated
prediction, one single coded block flag (cbf) signals the information whether at least
one non-zero transform coefficient level is transmitted for the whole CU. When
this so-called
rqt_root_cbf
is equal to 1, the RQT structure is signaled as
described above. Otherwise, no further residual information is transmitted and all
transform coefficient levels are inferred to be equal to zero. In the latter case, all
residual sample values are also equal to zero and no RQT syntax is transmitted.
The syntax element
rqt_root_cbf
is especially useful for coding of video
signals at low bit rates or for coding picture areas that can be motion-compensated
predicted in a sufficiently accurate way, because, with one single flag, a potentially
large number of transform coefficient levels equal to zero can be signaled very
efficiently. Also note that in the case of a skipped CU, i.e., a motion-compensated
CU with
cu_skip_flag
set equal to 1, no residual and hence, no transform
syntax including the
rqt_root_cbf
is transmitted. For intra-predicted CUs,
however, the
rqt_root_cbf
is always inferred to be equal to 1, since in that case,
a reasonable assumption is that at least some of the residual transform coefficient
levels are not equal to zero.
Furthermore, in case of
rqt_root_cbf
D
1, an additional cbf is transmitted
for each luma TB and each of the two associated chroma TBs. The significance
for luma TBs is signaled at the RQT leaf node level using the syntax element
cbf_luma
, while for the chroma components, the flags
cbf_cb
and
cbf_cr
are coded interleaved with the
split_transform_flag
symbols, i.e., at an
internal RQT node. This concept enables an efficient signaling for square blocks in
the video signal for which one or both of the chroma residual signals are equal to
zero, but the luma residual signal is not equal to zero. Under certain conditions, the
signaling of cbfs for luma and chroma TBs is also redundant and can be inferred
instead. Details on this redundancy reduction can be found in [
24
,
30
].
