Page 265 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

242 Chapter 10. Error Concealment Using Motion Field Interpolation

QSIF Table Tennis @ 30 f.p.s.
34
TR
AV
BM
33 MFI
BM-MFI
32
PSNR Y (dB) 31

30

29

28

27
10 20 30 40 50
Block loss rate (%)
Figure 10.7: Comparison between di,erent temporal concealment techniques when applied to
QSIF TABLE TENNIS at 30 frames=s. PSNRs are for damaged blocks only

In general, the best performance was achieved by BM-MFI, followed by
MFI, then BM, AV, and TR. As expected, TR performs well for the low-
movement AKIYO sequence. The poor performance of BM for AKIYO may be
due to an ambiguity problem where neighboring motion vectors give similar
SMD measures. A very interesting point to note is that the performance of
MFI starts to deteriorate for FOREMAN at high blockloss rates. This may be
due to the high dependency of MFI on the availability of the neighboring
motion vectors. This can be improved using interleaving techniques, as was
described in Chapter 9. In all cases, however, the BM-MFI technique main-
tained its superior performance. This is a clear indication of the robustness
of the technique. Over the three sequences and the considered blockloss rate
range, MFI provides on average 0:3 dB, 0:9 dB, and 1:4 dB improvements over
BM, AV, and TR, respectively, whereas BM-MFI provides a further 0:5dB
improvement over MFI. This corresponds to improvements of about 0:8 dB,
1:4 dB, and 1:9 dB over BM, AV, and TR, respectively.
th
Figure 10.8 shows the subjective quality of the 58 frame of TABLE TENNIS
with a blockloss rate of 30% when concealed using BM and BM-MFI. The
superior performance of the BM-MFI technique is immediately evident from
the good concealment of the left hand of the player. Note, however, that some
parts of the shirt are less sharp with the BM-MFI technique. This may be due
to the low-pass $ltering e,ect of the averaging (weighting) process.