Game Development Reference
In-Depth Information
where
s
2
=
+
log
2
N
indicates the number of bits to be shifted after the horizontal
transform. Accordingly, for the inverse horizontal transform process, the following
process is applied:
5
Y
1
=
(
X
×
T
N
+
(
1
<< (
s
3
−
1
))) >>
s
3
,
(5.18)
2
16
2
16
where
s
3
equals to 5. Then
Y
1
is clipped by the range of
, such that
Y
1
satisfies the 16-bit arithmetic limitation. For the inverse vertical transform process,
the following process is applied:
[−
,
−
1
]
T
N
Y
2
=
(
×
X
+
(
1
<< (
s
4
−
1
))) >>
s
4
,
(5.19)
where
s
4
equals to 21
−
n
, and
n
specifies the bit-depth of prediction residual.
5.3.2 Secondary Transform
For intraprediction block, a secondary transform can be used to further reduce the
correlation among the transform coefficients. The secondary transform matrix is
related with the block size. If the prediction block size is larger than 8
×
8 (including
8
4 secondary transform to the left-top corner of the transform block is
done, shown in Fig.
5.5
;for4
×
8), a 4
×
×
4 prediction block, a combined transform is used
instead of the traditional DCT, as shown in Eq.
5.20
.
Specifically, for an intra
M
1
×
M
2
block(
M
1
>
4 and
M
2
>
4), if the prediction
mode is 0
∼
2or13
∼
32, the secondary horizontal transform on the left-top 4
×
4
corner block
C
4
×
4
will be done as
S
4
.
Y
4
×
4
=
C
4
×
4
×
(5.20)
Fig. 5.5
Illustration of
secondary transform in
AV S 2
