Information Technology Reference
In-Depth Information
j
j
j
(
)
(
)
j
MAD k -
1
×
MAD k -
( )
MAD k -
1
MAD k
( )
k=
1
k=
1
k=
1
a=
1
2
j
j
(
)
2
j
(
MAD k -
1 )
-
MAD k -
(
1)
(7)
k=
1
k=
1
j
j
j
(
)
2
(
)
(
)
(
)
MAD k -
1
×
MAD k -
( )
MAD k -
1
MAD k MAD k -
( )
1
a=
k=1
k=
1
k=
1
2
2
j
j
(
)
2
j
(
MAD k -
1 )
-
MAD k -
(
1)
k=
1
k=
1
where j is the number of the encoded frames. Relative complexity of encoding frame
(RMAD) can be represented as the ratio of the predicted MAD of PMAD and AMAD,
and computed by
PMAD j
()
RMAD j =
()
. (8)
AMAD j
()
RMAD is a simple and accurate measure of frame complexity, and provides a mecha-
nism to control the target bits estimation. We quantize RMAD with a non-linear
strategy, C ( j ), as follows
0.5
RMAD
0.8
0.6
RMAD
0.8
< RMAD
1.0
Cj =
()
(9)
0.7
RMAD
1.0
< RMAD
RMAD >
1.8
1.8
1.8
Meanwhile, the importance of frame should be considered. Just as in JVT-G012 [5], it
deemed that P frame is more important than B frame. It should allocate more bits to P
frame. Similarly, a latter P frame is predicted from the former P frames and frames in
a GOP may have similar content. Hence, the higher quality the referenced frames are,
the smaller different between the referenced frames and the predicted frame will
probably be. Thus a video can get a higher quality at same cost of bandwidth. It can
be deemed that the distance of each P frame from the initial I frame in a GOP should
be considered when allocating bit. Parameter I ( j ) that denotes the importance of frame
is given as
R
( )
j-
PN
Ij=
()
(10)
N
p
where R PN ( j -1) is the number of P frames remaining for encoding, N p is total P frame
in a GOP. From these analyses, a new parameter CI ( j ) is defined as frame's complex-
ity and importance (CI), and calculated by
ζ (11)
where ζ is a constant, ranging from 1/9 to 1. The parameter CI ( j ) provides a new
measure for global encoding complexity. The equally distributed of the remaining bits
and buffer level to all non-coded frames lead to fluctuations in picture quality and
decrease in coding efficiency.
CI
()
j = C
()
j +
I
()
j
 
Search WWH ::




Custom Search