Compression - 3DTV: Processing and Transmission of 3D Video Signals - page 73

information, without the large amount of motion vector bits that would be
required in AVC.
Intra CUs may only use 2 N N partitions. For Intra CUs, PUs

describe the Intra prediction mode used in a particular area, rather than a
Motion Vector. HEVC defines 35 different intra-frame prediction modes for
improved intra-frame coding performance [8]. Combined with the flexible
tree-based partitioning, this gives a significant compression efficiency gain
(around 25%) compared to AVC Intra coding. The strong performance of the
Intra coding tools has led to the definition of a 'Still Image' profile, allowing
HEVC to be used for image compression.
HEVC allows different size Transform Units (TUs), which means that
different size integer transform sizes are used, from 4

× 2 N and N ×

32.
Each CU contains a transform tree, allowing different size transforms to
be used within a CU. Note that this also means that a transform must be
equal to or smaller than the size of the CU. Due to their larger size, and
larger multipliers, HEVCs transforms are significantly more complex for
some implementations than AVC.
Transform skip is another interesting new feature in HEVC. Transform
skip involves all of the normal processing for Intra- and Inter-coded CUs, but
allows the transform process to be bypassed. This feature may only be used
with 4

× 4upto32 ×

4 transforms. It seems to have advantages for text-based content,
and allows ringing around text to be significantly reduced. For coding of
screen-based content (e.g. PowerPoint presentations), the gain in quality is
very significant (

×

11%).
Compared to AVC, significant changes have been made to the loop
filtering. Similar to AVC, HEVC features an in-loop deblocking filter, which
has been refined and improved for inclusion in the new standard. HEVC
also features an additional loop filter: Sample Adaptive Offset (SAO) [9].
SAO applies simple pixel offsets to the reconstructed picture, to provide a
reference frame that more closely matches the original picture. An efficient
syntax for encoding the pixel offsets has been developed as part of the
standard. An encoder can specify different SAO offsets for each CTU.
HEVC uses a context-adaptive binary arithmetic coding (CABAC) algo-
rithm that is similar to the CABAC scheme deployed in the AVC standard.
On the other hand, unlike in AVC, CABAC is the only entropy encoder
scheme that can be used in HEVC. AVC allows context-adaptive variable
length coding (CAVLC) and CABAC to be two options at the entropy coding
stage. Context modelling is handled more efficiently in HEVC's CABAC
scheme that results in significant improvements in the throughput. Different
from the binary arithmetic coding method applied in AVC, the bypass-mode
within CABAC has also been improved in HEVC to further increase the
throughput [10].
The motion compensation filters of HEVC also differ from the filters
employed in AVC, where either half-sample or quarter-sample precision
filters with eight taps can be selected. AVC uses only a half-sample precision

>