Game Development Reference
In-Depth Information
decreased to improve quality. Specifically, the method can is shown as below,
If
(
img
>
tr % g op size !=
0
)
(9.19)
ʻ = ʻ
max
(
2
.
00
,
min
(
4
.
00
,
qp
/
8
.
0
))
const 2
(9.20)
where the const2 equals 0.6 under LDP common test conditions and 1.0 under RA
common test conditions.
9.1.4 Perceptual Coding Based on Divisive Normalization
The DCT domain SSIM index can be calculated as follows:
1
N 1
k
1
2
(
X
(
k
)
Y
(
k
))
=
1
2
(
X
(
0
)
Y
(
0
))
SSIM
(
x
,
y
) =
×
N
1
(9.21)
N
2
2
1
k = 1 ( X ( k )
X
(
0
)
+
Y
(
0
)
+
N
·
C 1
2
2
+ Y ( k )
)
+
C 2
N
1
where X
represent the DCT coefficients of the input signals x and y ,
respectively. C 1 and C 2 are used to avoid instability when the means and variances
are close to zero and N denotes the block size. The DCT domain SSIM index is
composed of the product of two terms, which are the normalized squared errors of
DC and AC coefficients, respectively. Moreover, the normalization is conceptually
consistent with the light adaptation (also called luminance masking) and contrast
masking effect of the HVS (Watson and Solomon 1997 ; Simoncelli and Heeger
1998 ). Equation ( 9.21 ) can be rewritten as
(
k
)
and Y
(
k
)
1
2
1
2
X ( 0 )
X ( k )
N
k = 1
1
)
ʷ dc
Y
(
0
1
)
ʷ ac
Y
(
k
SSIM
(
x
,
y
) =
ʷ dc
×
ʷ ac
N
1
(9.22)
where
2
2
ʷ dc =
X
(
0
)
+
Y
(
0
)
+
N
·
C 1 ,
(9.23)
N 1
k
2
2
1 (
X
(
k
)
+
Y
(
k
)
)
=
ʷ ac =
+
C 2 .
(9.24)
N
1
Equation ( 9.22 ) suggests that the DCT domain SSIM index can be computed from
normalized MSE of DC and AC coefficients. This inspires us to use SSIM-based
divisive normalization for perceptual video coding.
The divisive normalization factor is spatially adaptive and dependent on the con-
tent of the MB and determines the relative perceptual importance of each MB. The
MBs which are less important are quantized more coarsely as compared to the more
important MBs. The expected values of DC and AC energy are used as the reference
point to determine the importance of each MB. The MBs with higher energy than the
 
 
Search WWH ::




Custom Search