Page 262 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 10.4. Simulation Results 239

Hereafter, the term isolated error environment will be used to refer to this set
of test conditions.
All results in this subsection were generated using a full-search block-
matching algorithm with blocks of 16 × 16 pels, a maximum allowed dis-
placement of ± 15 pels, SAD as the distortion measure, restricted motion
vectors, and full-pel accuracy. Blocklosses were introduced randomly. Five
temporal error concealment techniques were simulated: temporal replacement
(TR), average vector (AV), boundary matching with side-match distortion
(BM), motion $eld interpolation (MFI), and the combination of BM and
MFI (BM-MFI). In each technique, the motion vectors of the four neigh-
boring blocks—left, right, above and below—were used in the concealment
displacement estimation stage. Whenever a neighboring motion vector was not
available, e.g., damaged or does not exist as in border blocks, it was set to
(0; 0). For the BM technique, SAD was used in the side-match distortion cal-
culations. Again, to maskany external e,ects, all quoted PSNRs in this set of
simulations were calculated for concealed blocks only and averaged over the
whole sequence. All quoted results refer to the luma components of sequences.

10.4.1.1 Choice of Parameters
Before evaluating the performance of MFI and BM-MFI, suitable values for
the smoothness parameters
and need to be chosen. Figure 10.3 shows the
e,ect of changing the smoothness parameter
on the performance of MFI
when applied to FOREMAN at 25 frames=s with di,erent blockloss rates. In
general, the performance is not particularly sensitive to the choice of
(a
change of about 0:3 dB). As
increases, the performance of MFI deteriorates
slightly. The best performance is achieved with
=1. This is approximately a
linear kernel. Thus, a linear interpolation kernel will be used in all subsequent
simulations. Note that a linear kernel also facilitates the use of a line-scanning
technique to reduce complexity, as was shown in Section 10.2.3.
Figure 10.4 shows the e,ect of changing the smoothness parameter on the
performance of BM-MFI when applied to FOREMAN at 25 frames=s with di,er-
ent blockloss rates. Again, the performance is not very sensitive to changes
in .As increases, the performance of BM-MFI slightly deteriorates. The
best performance is achieved with =1. The corresponding multihypothesis
weights are those shown in Figures 10.2(a) and 10.2(b). In what follows, this
value of will be used.

10.4.1.2 Performance Evaluation
Figures 10.5, 10.6, and 10.7 compare the performance of the $ve techniques
when applied to AKIYO,FOREMAN, and TABLE TENNIS, respectively. All results
were generated with a frame skip of 1.