Page 98 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media

ITU Journal: ICT Discoveries, Vol. 3(1), June 2020

VVC and EVC inter-coding tools are very close, one 2.4 Transforms
important difference being the VVC support of the
geometric partitioning mode. HEVC transforms are square separable NxN DCT-2
(Discrete Cosine Transform) for 4x4 to 32x32 block
Like HEVC, AV1 has three inter prediction types sizes, plus DST-7 (Discrete Sine Transform) for the
AMVP, Merge and Skip. The motion model can be intra 4x4 block size. The recent coding schemes
translational, affine, or global affine, with new introduce more variety with the support of multiple
predictions such as Wedge predictions, similar to separable transform types for square and
Geometric Partition Merge (GPM), and compound rectangular blocks, and for larger sizes up to 64x64.
inter-intra prediction similar to CIIP. Overlapped
Block Motion Compensation (OBMC) is used which In VVC and EVC, the concept of Multiple Transform
is not the case for the other solutions. Selection (MTS) is specified for residual coding of
both inter and intra-coded blocks, using DCT-2,
VVC and EVC include a motion refinement tool DCT-8 or DST-7 for square or rectangular blocks. In
named Decoder-side MV Refinement (DMVR). In addition, VVC inserts a set of Low Frequency Non-
addition, VVC defines two new prediction Separable Transforms (LFNST) implemented at the
refinement modes based on the optical flow, named encoder between the primary separable transforms
Bidirectional Optical Flow (BDOF) and Prediction and the quantization, while at the decoder between
Refinement with Optical Flow (PROF). the inverse quantization and the inverse primary
transform.
AV1 has also a richer set of square or rectangular
transforms: DCT-2, ADST (Asymmetric Discrete
Sine Transform), flipped ADST (applying ADST in
reverse order) and identity transform which is
equivalent to transform skip of HEVC or VVC.
The concept of RQT (Residual QuadTree) supported
in HEVC is not specified anymore in the three more
recent standards. Nevertheless, in VVC, a CU can be
split in smaller TUs using the Sub-Block Transform
(SBT).

Intra AMVP Merge Skip 2.5 In-loop filters
Fig. 3 – HEVC coding modes per coding unit New in-loop filters have improved the objective and
subjective performance of the new video coding
specifications. In VVC and EVC a new in-loop filter
called ALF (Adaptive Loop Filter) is inserted on top
of the Deblocking Filter (DF) and of the Sample
Adaptive Offset (SAO) filter used in HEVC. ALF is a
block-based adaption filter. For the luma
component, one filter among 25 is selected for each
4×4 block, based on the direction and activity of
local gradients. For chroma the choice is among
8 filters. Furthermore, a Cross-Component ALF is
introduced to refine chroma details lost in the
coding loop, by using co-located luma samples.
“Adaptive” means the filters can vary in a video
Intra Merge CIIP Skip Regular stream according to statististics of the content, and
Merge Affine Skip GEO also according to the blocks gradients-based
AMVP Merge GEO Skip Affine classification.
AMVP Affine Merge Regular Skip MMVD
Merge MMVD Skip ATMVP In VVC, a specific coding tool called luma mapping
Merge ATMVP with chroma scaling (LMCS) is added as a new
Fig. 4 – VVC coding modes per coding unit processing block prior to the loop filters. The luma
mapping is based on a piecewise linear model which

76 © International Telecommunication Union, 2020