Game Development Reference
In-Depth Information
Here, we first denote the past symbols of L i by
L i 1 ,
L i 2 ,...,
L 0 ,
if 1
i
<
N
L i 1
=
(6.1)
,
if
i
=
0
.
Then the context modeling for ( L i , R i ) can be defined as
L i 1
C
) (
) =
j if
(
Th
[
j
+
1
] >
Lmax
Th
[
j
] ),
(6.2)
(
L
,
R
L 1
with C
is equal to 0. L max is set to 0 at the beginning of decoding one
block, and the threshold array is
) (
)
(
L
,
R
0
,
1
,
2
,
3
,
5
,
8
,
11
, ,
intra _ luma
[
...
]=
0
,
1
,
2
,
3
,
4
,
7
,
10
, ,
inter _ luma
.
Th
0
7
(6.3)
0
,
1
,
2
,
3
,
5
, , , ,
chroma
Here the values of C ( L , R ) (g)
are the indices of the contexts, i.e., the 2D-VLC table
indices. According to Eqs. 8.2 and 8.3 , the value set of C ( L , R ) (g)
is {0, 1,
...
,6}for
luminance component and {0, 1,
, 4} for chrominance component. Each of them
corresponds to one different context, which corresponds to an interval I k derived
from Th
...
[
...
]
[
]
0
7
. The lower bound of I k is equal to Th
k
while the upper bound is
Th
. For example, for the luminance part of intramode, I 0 is equal to [0], and
I 3 is equal to [3, 5], and I 6 is [11,
[
k
+
1
]
].
The table switch process can be seen as a recursive context transition process,
which can be described as
L i 2
j
,
if
(
abs
(
L i 1 )
Th
[
C
) (
) +
1
]
(
L
,
R
L i 1
C
) (
) =
and
(
Th
[
j
+
1
] >
abs
(
L i 1 )
Th
[
j
] ).
(
L
,
R
L i 2
L i 2
]
(6.4)
Equation 6.4 actually describes that an appearance of a Level indicated by L i 1 ,
if its magnitude is equal or larger than the current used the upper bound of I k or
lower bound of I k + 1 , i.e., Th
C
) (
),
if
abs
(
L i 1 )<
Th
[
C
) (
) +
1
(
L
,
R
(
L
,
R
[
k
+
1
]
, triggers the context transition from one context
L i 2
L i 1
C
) (
)
to another C
) (
)
. That is, the prior coded abs
(
L i 1 )
is larger
(
L
,
R
(
L
,
R
L i 2
than the upper bound of C
's corresponding interval. The new context
is determined from the maximum previously coded magnitude among L i 1 . Such
context transition can exploit the increasing trend in magnitude from L 0 to L N 1
as well as the sequential dependency. Moreover, the use of the maximum values
of previously coded absLevels instead of the nearby coded Level can effectively
deal with those coefficients which are not in a monotonously increasing direction.
In this way, the drawback of the traditional VLC coding with one single VLC table
in one DCT block, which cannot adapt to locally statistical variations leading to low
efficiency, can be solved efficiently.
) (
)
(
L
,
R
Search WWH ::




Custom Search