Page 132 - 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

3. KEY TECHNICAL FEATURES 3.5 Parallelization
LCEVC deploys a small number of very specialized The scheme does not use any inter-block prediction.
coding tools that are well suited for the type of data The image is processed by applying small (2x2 or
it processes. Some of the key technical features are 4x4) independent transform kernels over the layers
highlighted below. of residual data. Since no prediction is made
between blocks, each 2x2 or 4x4 block can be
3.1 Sparse residual data processing
processed independently and in a parallel manner.
As further shown in Section 5, the coding scheme Moreover, each layer is processed separately, thus
processes one or two layers of residual data. This allowing the decoding of the blocks and decoding of
residual data is produced by taking differences the layers to be done in a largely parallel manner.
between a reference video frame (e.g., a source
video) and a base-decoded upscaled version of the 4. BITSTREAM STRUCTURE
video. The resulting residual data is sparse The LCEVC bitstream contains a base layer, which
information, typically edges, dots and details which may be at a lower resolution, and an enhancement
are then efficiently processed using very simple and layer consisting of up to two sub-layers. The
small transforms which are designed to deal with following section briefly explains the structure of
sparse information.
this bitstream and how the information can be
3.2 Efficient use of existing codecs extracted.
The base codec is typically used at a lower While the base layer can be created using any video
resolution. Because of this, the base codec operates encoder and is not specified further in the LCEVC
on a smaller number of pixels, thus allowing the standard, the enhancement layer must follow the
codec to use less power, operate at a lower structure as specified. Similar to other MPEG codecs
quantization parameter (QP) and use tools in a [3][4], the syntax elements are encapsulated in
more efficient manner. network abstraction layer (NAL) units which also
help synchronize the enhancement layer
3.3 Resilient and adaptive coding process information with the base layer decoded
The scheme allows the overall coding process to be information. Depending on the position of the frame
resilient to the typical coding artefacts introduced within a group of pictures (GOP), additional data
by traditional discrete cosine transform (DCT) specifying the global configuration and for
block-based codecs. The first enhancement controlling the decoder may be present.
sub-layer (L-1 residuals) enables us to correct The data of one enhancement picture is encoded
artefacts introduced by the base codec, whereas the into several chunks. These data chunks are
second enhancement sub-layer (L-2 residuals) hierarchically organized as shown in Fig. 1. For each
enables us to add details and sharpness to the processed plane (nPlanes), up to two enhancement
corrected upscaled base for maximum fidelity sub-layers (nLevels) are extracted. Each of them
(up to lossless coding). Typically, the worse the again unfolds into numerous coefficient groups of
base reconstruction is, the more the first layer may
contribute to correct. Conversely, the better the
base reconstruction is, the more bit rate can be Coefficient Group-1 data
(0)
allocated to the second sub-layer to add the finest
details. nLayers
3.4 Agnostic base enhancement Coefficient Group-1 data
(nLayers - 1)
The scheme can enhance any base codec, from Plane Y nLevels
existing ones (MPEG-2, VP8, AVC, HEVC, VP9, AV1,
etc.) to future and under-development ones Coefficient Group-2 data
(including EVC and VVC). The reason is that the nPlanes (0)
enhancement operates on a decoded version of the nLayers
base codec in the pixel domain, and therefore it can Plane V Coefficient Group-2 data
be used on any format as it does not require any (nLayers - 1)
information on how the base has been encoded
and/or decoded. Fig. 1 – Encoded enhancement picture data chunk structure

110 © International Telecommunication Union, 2020