IEEE Consumer Electronics Magazine - September/October 2022 - 14

Overview of VVC for Consumer Electronic Products
mechanisms is made implicitly ( i.e., without signaling)
and based on block characteristics.
Position-dependent prediction combination
(PDPC) is yet another new tool in VVC that
implicitly combines the prediction signal of a
block with its unfiltered and filtered boundaries.
Although position-dependent prediction combination
(PDPC) is applied differently on dc, planar,
horizontal/vertical, and other directional
modes, its general functionality can be formulated
as follows:
P~ðx; yÞ¼ bðvLRL þ vTRT
þð64 vL vTÞ ~Pðx; yÞþ 32Þ=26c
(1Þ
where ~Pðx; yÞ is the predicted sample at position
of coordinates ðx; yÞ, RT and RL are, respectively,
reference samples aligned from the top
and left to this position. Weights of these reference
samples, expressed as vT and vL, respectively,
are determined based on the position
ðx; yÞ as well as the selected intra-mode. Finally,
in this equation, the operator bxc is the lower
integer value of a real x used with a division by
26 to normalize the modified prediction value
into its range.
Intra-subpartitions(ISP) tool allows splitting
an intra-block into two or four sub-blocks, each
having its own separate residual block, while
sharing one single intra-mode. The initial motivation
behind this tool was to allow short-distance
intra-prediction of blocks with nonstationary
texture, by sequential processing of subpartitions.
On thin nonsquare blocks, this scheme
can result in subpartitions coded with 1D residual
blocks, providing the closest possible reference
lines for intra-prediction.
The matrix-based intra-prediction (MIP) in
VVC is a new tool designed by an AI-based datadriven
method. matrix-based Intra prediction
(MIP) modes replace the functionality of conventional
IPMs, by matrix multiplication of reference
lines, instead of their directional projection. The
process of predicting a block with MIP consists
in three main steps: 1) averaging with subsampling,
2) matrix-vector multiplication, and 3) linear
interpolation. The data-driven aspect of MIP
is expressed in the second step, where a set of
matrices are pretrained and hard-coded in the
VVC specification. This set is designed to
14
provide distinct matrices for different combinations
of block size and internal MIP mode.
Cross-component linear model is a new tool
in VVC for exploiting local correlations between
luma and chroma channels. This tool is based on
a similar concept in the HEVC range extension
standard, where the inter-channel correlation is
modeled in the residual domain. In VVC, this
modeling is carried out in the reconstructed
pixel domain, in the form of
P~cðx; yÞ¼ a ^Pðx; yÞþ b
(2)
where P~cðx; yÞ is the predicted chroma value at
position ðx; yÞ based on the co-located reconstructed
luma value
P^ðx; yÞ. The model parameters
a and b are explicitly derived based on the
relation between neighboring luma and chroma
samples.
Inter-Coding Tools
Inter-coding relies on inter-prediction of
motion and texture data, from previously reconstructed
pictures stored in the decoded picture
buffer. A simplified block diagram of the VVC
inter-prediction process is provided in Figure 3.
The process first involves motion prediction,
based on a list of motion data candidates. The
motion prediction can be corrected by residual
motion information signaled in the bitstream.
The reconstructed motion vectors are then used
to perform one or two motion compensations,
whether the coding block (CB) is uni- or bipredicted.
When biprediction is performed, a blending
process is then applied to mix the two
motion-compensated blocks. Finally, a prediction
enhancement step is performed as a postprediction
filtering. The resulting predicted
signal can be further corrected by a residual
block signaled in the bitstream.
Two different motion models are supported
in VVC, translational model and affine model,
controlled at the CU-level. The affine model can
rely on 4 or 6 degrees of freedom. When a CU is
coded with affine motion, it is split into 44 subblocks,
whose motion vectors (MVs) are derived
from the affine parameters of the CU. These
parameters are deduced from two or three control
point motion vectors positioned in the topleft,
top-right, and possibly bottom left corners
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - September/October 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2022
