Page 268 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 10.4. Simulation Results 245

QSIF Foreman at 24 kbits/s
28
TR
AV
27 BM
MFI
BM-MFI
26
25
PSNR (dB) 24

23

22

21
20
10 –4 10 –3
BER
Figure 10.9: Comparison between di,erent temporal concealment techniques when applied to
QSIF FOREMAN. The sequence was H.263 encoded at 24 kbits=s and then corrupted with a range
of bit error rates

represents the PSNR of the three components together with a 4:2:0 subsam-
pling. Again, the best performance in each case was achieved by BM-MFI,
followed by MFI. For example, for the TABLE TENNIS sequence, MFI provides
improvements of 0:27 dB, 0:54 dB, and 0:31 dB over BM, AV, and TR, respec-
tively, whereas BM-MFI provides a further 0:24 dB improvement over MFI.
This corresponds to improvements of about 0:51 dB, 0:78 dB, and 0:55 dB over
BM, AV, and TR, respectively.
Figure 10.9 shows the performance of the $ve techniques when used to
conceal the 24 kbits=s QSIF FOREMAN sequence corrupted with BERs in the
range 10 −4 to 10 −3 . At low BERs the di,erences between the techniques are
small. However, as the BER increases, the techniques split into three perfor-
mance levels. The lowest level includes TR and AV, the next level includes
BM and MFI, and the highest level includes BM-MFI.
Figure 10.10 shows a frame of the 24 kbits=s QSIF FOREMAN corrupted with
a BER of 10 −3 and then decoded and concealed using BM and BM-MFI.
The superior performance of the BM-MFI technique is immediately evident,
especially at the eyes and the edges of the face.
It is worth noting here that at a BER of 10 −3 , the PSNRs of the con-
cealed sequences drop by about 5–9 dB compared to the error-free values,
and the subjective quality may not be acceptable. A close inspection of the