Graphics Reference
In-Depth Information
syntax elements to reduce the side information and speed-up SAO processing by
achieving more efficient memory access on certain platforms [
4
]. These syntax
elements are therefore only coded for the Cb component. The design of the
codewords (including “off”, “EO class selection index”, and “BO band position
index”) is based on the probability distribution to reduce side information.
7.3.4.3
CABAC Contexts and Bypass Coding
All CTU-level, SAO syntax elements including SAO merging information, SAO
type information, and offset information are coded with context-based adaptive
binary arithmetic coding (CABAC). Only the first bin of the SAO type, which
specifies whether SAO is turned on or off in the current CTU, and the SAO merge-
left and merge-up flags use CABAC contexts. All other bins are coded in the bypass
mode, which significantly increases the SAO parsing throughput in CABAC without
much coding efficiency loss [
1
,
3
,
15
,
30
,
45
].
7.4
Implementation and Parallelization Aspects
7.4.1
Deblocking Filter Complexity and Parallelism
When designing the HEVC deblocking filter, a lot of attention was paid to complex-
ity and parallelization aspects. In H.264/AVC video decoders, deblocking takes a
significant part of the computational complexity [
27
,
28
]. Moreover, in H.264/AVC,
the deblocking operations at one block boundary may affect the samples used in
deblocking of the next block boundary, which complicates parallel processing.
7.4.1.1
HEVC Deblocking Filter Complexity
The complexity of the HEVC deblocking filter has been significantly decreased
compared to the H.264/AVC deblocking. First of all, the deblocking is only applied
to the block boundaries on the 8
8 luma sample grid. This already decreases
the worst-case complexity of the deblocking compared to H.264/AVC, where the
deblocking is applied on the 4
4 sample grid. The average complexity of the
deblocking operation is also decreased compared to H.264/AVC since the prediction
and transform blocks in HEVC are on average larger than those in H.264/AVC,
where the maximum size of prediction blocks is 16
16 luma samples and the size
of the transform blocks is 8
8 luma samples (if the maximum transform size in
HEVC is not restricted to the same limits resulting in the worst-case scenario).
The filtering decisions constitute a significant part of the deblocking filter
complexity. In order to reduce the complexity of deblocking decisions, the HEVC
