Databases Reference
In-Depth Information
QA B C D E F G H
8
I
abcd
1
J
e
f
gh
K
i
j
kl
6
L
m
nop
3
4
7
5
0
F I GU R E 19 . 17
Prediction modes for 4
×
4 intra prediction.
label is given by
|
θ
i
,
j
|
α
i
,
j
(
Q
step
)
Q
step
l
i
,
j
=
sign
(θ
i
,
j
)
In order to broaden the quantization interval around the origin we add a small value in the
numerator:
|
θ
i
,
j
|
α
i
,
j
(
Q
step
)
+
f
(
Q
step
)
l
i
,
j
=
sign
(θ
i
,
j
)
Q
step
In actual implementation we do away with divisions and the quantization is implemented as
[
262
]
f
2
17
+
Q
E
l
i
,
j
=
sign
(θ
i
,
j
)
[|
θ
i
,
j
|
M
(
Q
M
,
r
)
+
]
>>
17
+
Q
E
where
Q
M
=
Q
step
(
mod
6
)
Q
step
6
Q
E
=
⎧
⎨
,
0
i
j
even
,
r
=
1
i
j
odd
⎩
2
otherwise
>>
denotes a binary right-shift, and
M
is given in Table
19.9
.
The inverse quantization is given by
θ
i
,
j
=
(
Q
M
,
)
l
i
,
j
S
r
Q
E
where
denotes a left-shift, and
S
is given in Table
19.10
Prior to quantization, the transforms of the 16
<<
8
chrominance residuals of themacroblock-based intra prediction are processed to further remove
redundancy. Recall that macroblock-based prediction is used in smooth regions of the
I
picture. Therefore, it is very likely that the DC coefficients of the 4
×
16 luminance residuals and the 8
×
4 transforms are heavily
correlated. To remove this redundancy, a discrete Walsh-Hadamard transform is used on the
×
