Game Development Reference
In-Depth Information
and organize the encoding data (Yu et al. 2012 ). First, pictures are split into largest
coding units (LCUs), which consist of 2N
×
2N samples of luminance component
and associated chrominance samples with N
8, 16 or 32. One LCU can be a single
CU or can be split into four smaller CUs with a quad-tree partition structure; a CU can
be recursively split until it reaches the smallest CU size limit, as shown in Fig. 3.2 a.
Once the splitting of the CU hierarchical tree is finished, the leaf node CUs can be
further split into PUs. PU is the basic unit for intra- and inter-prediction and allows
multiple different shapes to encode irregular image patterns, as shown in Fig. 3.2 b.
Size of PU is limited to that of CUwith various square or rectangular shapes. Besides
CU and PU, TU is also defined to represent the basic unit for transform coding and
quantization. The size of a TU cannot exceed that of a CU, but it is independent
of the PU size. After transformation, context adaptive arithmetic coding is used for
the entropy coding of the transformed coefficients. A two-level coefficient scan and
coding method can encode the coefficients of large blocks more efficiently.
=
3.2 Coding Tools
The tools used in AVS2 can be classified into four categories: prediction coding
including intra prediction and inter prediction, transform, entropy coding and in-
loop filtering, and tools for scene video coding. We will give a brief introduction to
the coding tools in this section.
3.2.1 Intra Prediction
Intra prediction is used to reduce the redundancy existing in the spatial domain of
the picture. Block partition-based directional prediction is used for AVS2 (Piao et al.
2014 ). As shown in Fig. 3.2 , besides the square PU partitions, non-square partitions,
called short distance intra prediction (SDIP), are adopted by AVS2 for more efficient
intra luminance prediction (Yu et al. 2013 ), where the nearest reconstructed boundary
pixels are used as the reference sample in intra prediction. For SDIP, a 2N
2N PU
is horizontally/vertically partitioned into 4 prediction blocks. SDIP is more adaptive
to the image content, especially in edge area. But for the complexity reduction,
SDIP is used in all CU sizes except 64
×
64 CU. For each prediction block in the
partition modes, totally 34 prediction modes are supported for luminance, including
31 angular modes (Piao et al. 2014 ), plane mode, bilinear mode, and DC mode. As
illustrated in Fig. 3.3 , the prediction directions associated with the 31 angular modes
are distributed within the range of [
×
135 ,45 ]. Each sample in a PU is predicted
by projecting its location to the reference pixels applying the selected prediction
direction. To improve the intra prediction accuracy, the subpixel precision reference
samples must be interpolated if the projected reference samples locate on a non-
integer position. The non-integer position is bounded to 1/32 sample precision to
 
Search WWH ::




Custom Search