Digital Signal Processing Reference
In-Depth Information
Fig. 5.25
Zero-padding
optimization process
start
Input
f,
initialize
N
Find a smallest
N
to
satisfy Condition 1 and 2
N
Is
D
1
almost
invariable?
Y
End
robustness comparison over packet-loss channel. In all of these experiments, the
luminance components of standard sequences
Fo re m a n
and
Hall
in QCIF@15 fps
(frame per second) are tested. The total bit rate of two channels is accounted.
5.6.5.1
Comparison of Different DVC Methods
The performance of single description DVC (SD DVC) used will affect the
performance of MDDVC proposed, so firstly, we compare our proposed SD HDVC
with other SD DVCs referenced, including DVC of [
21
] and residual DVC of [
18
].
The GOP varies in 2, 4, and 8 to show the performance comparison.
Comparison curves in Fig.
5.26
shows that the proposed SD HDVC obtains up to
3 dB improvement for
Hall
sequence. There is even up to 3.9 dB improvement
more than residual DVC in [
18
]. These improvements originate from the fact
that the hybrid scheme can make use of some temporal and spatial correlations
simultaneously at the encoder. For
Fo re m a n
sequence, the improvement can also be
achieved, although it is not as valuable as the
Hall
sequence due to the low temporal
correlation.
Then, we plot the performance of the proposed MDDVC in Figs.
5.27
and
5.28
,
where the MD channels are assumed to be ideal, in which one channel is either intact
or totally lost, that is, the information loss rate is 0% or 50%. The results show that
the number of zeros added will affect the results of MDDVC, and even show that
more zeros padded are not always helpful for the side decoder of MDDVC, such
as in
Hall
sequence with 176 zeros. So an optimized zero-padding is necessary for
introducing the proper number of zeros in the preprocessing stage.
