Page 275 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

252 Chapter 10. Error Concealment Using Motion Field Interpolation

Table 10.4: Spatial-temporal recovery for QSIF AKIYO with M =10, QP =10, skip = 3, and a
macroblockerror rate of 30%
Spatial-components recovery
ZR AV BM MFI

Temporal- TR 27.91 27.80 26.79 29.22
component AV 27.23 27.49 27.07 28.58
recovery BM 28.33 28.10 27.38 29.48
MFI 27.42 27.58 26.38 28.69

Table 10.5: Spatial-temporal recovery for QSIF FOREMAN with M =10, QP =10, skip = 3, and a
macroblockerror rate of 30%

Spatial-components recovery
ZR AV BM MFI
Temporal- TR 18.57 19.68 20.32 20.71
component AV 18.25 19.58 19.72 20.56
recovery BM 18.59 20.13 20.80 21.18
MFI 18.14 19.51 19.65 20.48

temporal-component recovery. For example, in Figure 10.13(b), at a frame
skip of 3, moving from the best technique, ZR-BM, to the worst technique,
ZR-MFI, drops the quality by about 0:3 dB, whereas in Figure 10.16(b) mov-
ing from the best technique, MFI-ZR, to the worst technique, AV-ZR, drops
the quality by about 1 dB. It can be concluded also that spatial-components re-
covery is, in general, more complex than temporal-component recovery. With
temporal-component recovery, a simple technique like ZR can be suNcient,
whereas with spatial-components recovery more complex techniques like MFI
and BM are essential. Furthermore, the results of Sections 10.5.1 and 10.5.2
indicate that the combination MFI-BM (i.e., spatial recovery using MFI and
temporal recovery using BM) may provide the best spatial-temporal recovery.
This is con$rmed in Tables 10.4, 10.5, and 10.6, which show the performance
of all 16 possible combinations with a frame skip of 3 and a macroblock error
rate of 30%.

10.5.4 Multihypothesis Temporal Error Concealment

It was demonstrated in Section 10.4 that a more robust performance can
be achieved if the concealed blockis a weighted average of a number of