Graphics Reference
In-Depth Information
where
D
2
4=6
;2
3=6
;2
2=6
;2
1=6
;2
0
;2
1=6
T
G
D
ŒG
0
;G
1
;G
2
;G
3
;G
4
;G
5
T
The fixed point approximation of (
6.7
)inHEVCisgivenby
g
QP %6
D
round
2
6
G
QP %6
This results in
g
D
Œg
0
;g
1
;g
2
;g
3
;g
4
;g
5
T
D
Œ40; 45; 51; 57; 64; 72
T
HEVC supports frequency-dependent quantization by using quantization matri-
ces for all transform block sizes. Let
W
[
x
][
y
] denote the quantization matrix weight
for the transform coefficients at location (
x
,
y
) in a transform block. A value of
W
[
x
][
y
]
D
1 indicates that there is no weighting. The fixed point representation of
W
[
x
][
y
] is given by:
w
Œx Œy
D
round .16
W Œx Œy/
where
w
[
x
][
y
] is represented using 8-bit values.
For a quantizer output,
level
[
x
][
y
], the de-quantizer is specified in the HEVC
standard as
coeff
Q
Œx Œy
D
level
Œx Œy
w
Œx Œy
g
QP %6
<<
C
offse
t
IQ
QP
6
>>
shift1
(6.8)
250
200
150
100
50
Fig. 6.6
Relationship
between quantization step
size (
Qstep
) and quantization
parameter (
QP
)
0
0
5
10
15
20
25
30
35
40
45
50
QP