Information Technology Reference
In-Depth Information
un
Tn=
()
j
()
+ l n-Bn
γ
( ()
( )
, (3)
buf
j
j
c
j
F
r
γ
is a constant and its typical value is 0.75 [5]. Meanwhile, the remaining bits
are computed by
where
Tn
Tn=
R
()
r
j
()
, (4)
ref
j
(1)
j-
PN
PN
R j-
is the number of P frames remaining for encoding. The final target
bit
R
for the
j
th frame is calculated by
()
where
( )
R n= T n+
β
* ()(1-)* ()
β
T n
, (5)
j
ref
j
buf
j
where
β
is a weighting factor and set typically as 0.5 [5].
In Eq.(2), all frames have an equal number of target buffer level. In Eq.(4), the
remaining bits
T
r
is also allocated to all non-coded frames equally. Thus, a buffer nearly
full will allocate less target bits to a new frame while a nearly empty buffer will allocate
more bits, which will lead to a much smaller quantization parameter regardless of the
complexity of frame content. Inaccurately estimate target bits for the current P frame
results in fluctuations in picture quality and decrease in coding efficiency.
In the proposed scheme, we focused on
T
r
and
Tbl
,
that is, the remaining bits and
buffer level should be un-equally distributed to all non-coded frames according to
frame complexities and importance in the target bit estimation step. In other words,
different complex and important frame will get different buffer and bandwidth
resource. Details of the improvements will be discussed in Section 3.
3 The Proposed Rate-Control Method Using Frame CI
The basic idea in this paper is to allocate more bits for scene change frames or high
complexity frames or for important frames, and less bits for low complexity frames or
unimportance frames to achieve constant quality. It is well known that MAD can be a
good indication of encoding complexity of the residual component. In the quadratic
rate-quantization (R-Q) model, the encoding complexity is usually substituted by
MAD [5]. Lee et al. measure 4x4 Intra-block complexity by using MAD with 5x5
statistical window [12]. Based on their contribution, we defined a new factor to
describe P frame parameter complexity and importance, denoted as CI, and proposed
a rate-control method using frame CI.
3.1 Complexity and Importance Measure of P Frame
Average actual MAD (AMAD) of all previously encoded P frames in GOP is defined
to represent the complexity of encoded P frames. AMAD is calculated as follows
1
1
j-
∑
AMAD j =
()
MAD k
()
, (6)
j
k=
1
Then, a linear prediction model, like in [5], is employed to calculate the predicted
MAD (PMAD) of the current frame by
. A method
similar as updating parameters of R-D model, like in MPEG-4[4], is given to update
a
1
and
a
2
as
PMAD j = a
()
×
AMAD j -
( 1)
+ a
1
2
Search WWH ::
Custom Search