Page 99 - 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

adjusts the dynamic range of the input signal by the same picture. In VVC the motion search area is
redistributing the codewords across the signal limited to the current CTU and part of the left CTU
range to improve compression efficiency. For the to control the complexity. Palette reduces the
chroma the scaling applies to the prediction number of codewords to encode to a limited
residual and depends on the average value of top number of triplets (one luma and two chroma
and/or left reconstructed neighbouring luma values). AV1 has equivalent modes to IBC and
samples. As a consequence, with LMCS all the Palette, but EVC has only IBC.
reconstruction processing in luma (inverse BDPCM and ACT have been reintroduced in VVC
quantization, inverse transformation and with some adaptations. BDPCM allows us to encode
prediction for inter and intra modes) is made in the prediction residuals without transform (Transform
mapped domain. Like ALF, the mapping function Skip) by predicting them from the previously coded
can vary in a video stream according to content residuals. BDPCM is relevant for Screen Content but
statistics.
also for lossless compression. ACT reduces the
AV1 also combines different in-loop filters in redundancy between color components in typically
addition to a regular deblocking filter. RGB coding through a color space conversion.
A Constrained Directional Enhancement Filter 2.8 Other tools
(CDEF), which de-rings contours and preserves the
details to be applied after deblocking, works by A new feature called Reference Picture Resampling
estimating edge directions. A loop restoration filter (RPR), which allows the picture size to vary from
can be applied selectively according to noise level. picture to picture, is specified in VVC. This feature is
also available in AV1 but in the horizontal direction
2.6 Entropy coding
only. RPR offers an additional level of flexibility for
Arithmetic coding is used in all the video coding bit-rate control to adapt to network bandwidth
solutions addressed in this paper. HEVC entropy variation.
coding was based on the so-called CABAC (Context- In HEVC, annexes were produced after the
based Adaptive Binary Arithmetic Coding) initially publication of the initial version of the standard to
introduced in AVC [1]. CABAC is composed of four include spatial and temporal scalability, as well as
main steps: the binarization of each syntax element the coding of multiview (3DTV) content. These
in a binary string, followed by the choice of a extensions were based on the layer concept, which
probability model for each bit (or “bin”) to be “0” or is available in the core specification of VVC.
“1” based on the context, then the binary arithmetic
coder encodes each bin with less than a bit, and Furthermore, VVC offers high-level syntax (HLS)
finally the probability of the selected model is features such as self-decodable sub-pictures for bit
updated. VVC has continued to improve CABAC with stream extraction and merge applications, or view-
a higher accuracy on the probabilities derivation, by dependent streaming. Gradual Decoding Refresh
increasing the number of probability models, and (GDR) is another HLS feature adapted to low delay
by updating the probabilities through an adaptive applications to avoid the burst of complete Intra
double-window. EVC entropy coding is also a pictures.
CABAC engine, but does not benefit from all the
enhancements brought in VVC. AV1 entropy coding 3. TEST CONDITIONS
is different as it uses a multi-symbol entropy coding The test conditions for the video coding evaluations
per syntax element without binarization step.
reported in this paper are derived from the JVET
2.7 Screen Content Coding Common Test Conditions (CTC) [8]. Two scenarios
are considered:
In its Screen Content Coding (SCC) extensions
issued in 2016 [7], HEVC specifies four new coding • Broadcast scenario with one Intra picture
tools adapted to “Screen Content” which are approximately every second,
pictures partially or totally composed of computer • Streaming scenario with one Intra picture
graphic objects. These tools are Intra Block Copy approximately every two seconds.
(IBC), Palette, Block-wise Differential Pulse Code The Broadcast scenario is the same as the “Random
Modulation (BPDCM) when Transform Skip is Access” case in JVET Common Test Conditions (CTC)
allowed, and Adaptive Color Transform (ACT). IBC [8]. It simulates a real scenario allowing a TV user
is equivalent to a motion compensation but within to zap from a channel to another at an acceptable

© International Telecommunication Union, 2020 77