Graphics Reference
In-Depth Information
Def
IntraL
f
L
ault 8x8 for
Luma, IntraCb,
IntraCr
Default 8x8 for
InterLuma, InterCb,
InterCr
L
Default 4x4 for
IntraLuma, IntraCb,
IntraCr, InterLuma,
InterCb, InterCr
16
16
16
1
1
16
17
18
21
2
4
16
16
16
16
17
18
20
24
16
16
16
1
1
16
17
19
22
2
5
16
16
16
17
18
20
24
25
16
1
6
16
16
16
16
17
1
1
18
20
22
25
2
9
16
16
17
18
20
24
25
28
16
1
6
16
16
16
16
18
2
2
21
24
27
31
3
6
16
17
18
20
24
25
28
33
17
17
20
2
4
30
35
41
4
7
17
18
20
24
25
28
33
41
16
1
6
16
16
18
19
22
2
7
35
44
54
6
6
5
18
20
24
25
28
33
41
54
16
1
6
16
16
21
22
25
3
3
31
41
54
70
88
20
24
25
28
33
41
54
71
(Flat matrix)
24
25
29
3
3
36
47
65
88
1
15
24
25
28
33
41
54
71
91
Fig. 6.7
Default quantization matrices for transform blocks of size 4
4and8
8
6.3.1
Quantization Matrix
In HEVC, the encoder can signal whether or not to use quantization matrices
enabling frequency dependent scaling. Frequency dependent scaling is useful to
carry out human visual system (HVS)-based quantization where low frequency
coefficients are quantized with a finer quantization step size when compared to
high frequency coefficients in the transform block [
12
]. HVS-based quantization
can provide better visual quality than frequency independent quantization on some
video sequences. HEVC uses the following 20 quantization matrices that depend on
the size and type of the transform block:
Luma:In ra4
4, Inter 4
4, Intra 8
8, Inter 8
8, Intra 16
16, Inter 16
16,
Intra 32
32, Inter 32
32
Cb:In ra4
4, Inter 4
4, Intra 8
8, Inter 8
8, Intra 16
16, Inter 16
16
Cr: Intra 4
4, Inter 4
4, Intra 8
8, Inter 8
8, Intra 16
16, Inter 16
16
When frequency dependent scaling is enabled by using the syntax element
scaling_list_enabled_flag
, the quantization matrices of sizes 4
4and
8
8 have default values as shown in Fig.
6.7
. The default quantization matrices
for transform blocks of size 16
16 and 32
32 are obtained from the default 8
8
quantization matrices of the same type by upsampling using replication as shown
in Fig.
6.8
. The red colored blocks in the figure indicate that a quantization matrix
entry in the 8
8 quantization matrix is replicated into a 2
2regioninthe16
16
quantization matrix and into a 4
4 region in the 32
32 quantization matrix. 8
8
matrices are used to represent 16
16 and 32
32 quantization matrices in order to
reduce the memory needed to store the quantization matrices.
Non-default quantization matrices can also be optionally transmitted in the
bitstream in sequence parameter sets (SPS) or picture parameter sets (PPS).
Quantization matrix entries are scanned using an up-right diagonal scan and
DPCM coded and transmitted. For 16
16 and 32
32 quantization matrices,
only size 8
8 matrices (which then get upsampled to the correct size in the
