Information Technology Reference
In-Depth Information
(3)
(4)
Where
r
x
and
r
y
are calculated from equations (3) and (4) depending of the MVs
halves (MV
x
/2 and MV
y
/2) provided by DVC or a minimum value of 2 to avoid
applying too small search ranges. Notice that each H.264 subpartition related to a
particular H.264 MB takes advance of the same area reduction.
4 Performance Evaluation
The source WZ video 8 was generated by VISNET II codec using a fixed matrix QP =
7 and GOPs 2, 4 and 8. While sequences (in QCIF format) are decoded, the MVs are
passed to the H.264 encoder without any increase of complexity. Afterwards, the
transcoder converts this WZ video input into a H.264 video stream using QP = 28, 32,
36, 40 as specified in
Bjøntegaard and Sullivan´s
common test rule [10][9]. Every
WZ GOP pattern was mapped into a H.264 GOP I11P. The simulations were run by
using the version JM 14.1 of H.264 and the baseline profile with the default
configuration file. The baseline profile was selected because it is the most used profile
in real-time applications due to its low complexity. For the same reason,
RDOptimization was turned off. For ME, search area was defined by a window with
32 pixel length. In order to check our proposal we have chosen four representative
sequences with different motion levels at 15 fps and 30 fps coding 150 frames and
300 respectively, the same sequences that were selected in the DISCOVER codec's
evaluation [8]. On the other hand, the
percentage of
Time Reduction
(%TR) reported
Table 1.
Performance of the proposed transcoder for 15fps sequences
RD performance of the WZ/H.264 video transcoder - 15fps
Sequence
GOP
Δ
PSNR (dB)
Δ
Bitrate (%)
TR (%)
2
-0.076
2.00
72.75
Foreman
4
-0.076
2.04
72.80
8
-0.073
2.04
73.49
2
-0.009
0.23
67.03
Hall
4
-0.007
0.16
66.92
8
-0.006
0.16
66.63
2
-0.057
1.51
77.97
CoastGuard
4
-0.050
1.35
77.84
8
-0.055
1.49
78.07
2
-0.145
4.11
68.53
Soccer
4
-0.150
4.45
69.94
8
-0.148
4.66
70.11
mean
-0.071
2.02
71.84
Search WWH ::
Custom Search